@mindstudio-ai/agent 0.1.139 → 0.1.141

package/dist/cli.js

@@ -1433,7 +1433,6 @@ function getOrOpen(url, token) {
       pending: /* @__PURE__ */ new Map(),
       chunked: false,
       inbound: /* @__PURE__ */ new Map(),
-      acksSeen: false,
       retired: false
     };
     const c = self;
@@ -1457,7 +1456,6 @@ function getOrOpen(url, token) {
         return;
       }
       if (msg.t === "db-ack") {
-        c.acksSeen = true;
         const p = c.pending.get(msg.id);
         if (p && !p.acked) {
           p.acked = true;
@@ -1533,15 +1531,11 @@ async function executeDbBatchOverWs(url, token, databaseId, queries) {
   }
   return await new Promise((resolve, reject) => {
     const pending = { resolve, reject, timer: null, acked: false };
-    if (self.acksSeen) {
-      pending.timer = setTimeout(() => {
-        forget(self, id);
-        retire(self, "a request went unacknowledged");
-        reject(new DbWsTransportError("ws ack timed out", true));
-      }, WS_ACK_TIMEOUT_MS);
-    } else {
-      armIdleTimer(self, id, pending);
-    }
+    pending.timer = setTimeout(() => {
+      forget(self, id);
+      retire(self, "a request went unacknowledged");
+      reject(new DbWsTransportError("ws ack timed out", true));
+    }, WS_ACK_TIMEOUT_MS);
     self.pending.set(id, pending);
     let started = false;
     let complete = false;
@@ -4569,7 +4563,8 @@ var init_source = __esm({
        * it is who or what was written about then. No filter covers the corpus.
        * `chunks` is how many matching passages carry each id, not documents: a
        * long document's passages each count. Approximate counts, ordered most
-       * first; `limit` defaults to 20, at most 100.
+       * first; `limit` defaults to 20, at most 1,000. One document's entities:
+       * `{ filter: { documentIds: [id] } }`.
        *
        * Shares the search rate limit (300/minute per app), and shows the same
        * `index_warming` / `index_building` as search while a corpus loads.
@@ -4717,7 +4712,15 @@ var init_source = __esm({
        * as the new one is registered, before its build lands; a search in that
        * window sees neither. **Throws `mapper_skipped`** (HTTP 422) when the
        * mapper skipped the object — a specific thing it refused, which the caller
-       * should see — and `mapper_error` when the mapper threw.
+       * should see — and `mapper_error` when the mapper threw, or when none of
+       * its documents passed validation. A document that fails is left out and
+       * reported in {@link AddResult.invalid}; the rest are registered.
+       *
+       * An object that maps to more than a thousand documents (a large export,
+       * streamed out of the mapper as it parses) registers its first thousand
+       * here and returns {@link AddResult.job}, a job that registers and builds
+       * the rest; `documentCount` says how many there are in all. Another job on
+       * the source refuses it with `data_source_busy`.
        */
       async add(content, options) {
         const bytes = typeof content === "string" ? Buffer.from(content) : Buffer.from(content);
@@ -4727,6 +4730,7 @@ var init_source = __esm({
           ...options.contentType ? { contentType: options.contentType } : {},
           ...options.metadata !== void 0 ? { metadata: options.metadata } : {},
           ...options.entities !== void 0 ? { entities: options.entities } : {},
+          ...options.budgetDollars !== void 0 ? { budgetDollars: options.budgetDollars } : {},
           body: bytes.toString("base64")
         });
       }
@@ -4760,6 +4764,59 @@ var init_source = __esm({
         });
         return { documents: documents ?? [], nextCursor: nextCursor ?? null };
       }
+      /**
+       * One page of the passages matching a filter, unranked: every page that
+       * mentions an asset (`{ entities: ['P-1004A'] }`), every passage holding a
+       * phrase, every chunk of some documents. Where {@link search} ranks and
+       * stops at its top results and {@link count} gives a number, this lists
+       * them all. Passages come document by document, each document's in order;
+       * pass `nextCursor` back for the next page, which is null after the last.
+       * `limit` defaults to 50, at most 200.
+       *
+       * Shares the search rate limit (300/minute per app), and shows the same
+       * `index_warming` / `index_building` as search while a corpus loads.
+       */
+      async passages(options) {
+        const { passages, nextCursor, latencyMs } = await this._call("passages", {
+          slug: this._slug,
+          ...options?.filter !== void 0 ? { filter: options.filter } : {},
+          ...options?.cursor ? { cursor: options.cursor } : {},
+          ...options?.limit !== void 0 ? { limit: options.limit } : {}
+        });
+        return {
+          passages: passages ?? [],
+          nextCursor: nextCursor ?? null,
+          latencyMs: latencyMs ?? 0
+        };
+      }
+      /**
+       * Every passage matching a filter, a page at a time behind the scenes: the
+       * walk over {@link passages}. Each page counts against the search rate
+       * limit, so a large walk belongs in a background task.
+       *
+       * ```ts
+       * for await (const passage of Manuals.allPassages({ entities: ['P-1004A'] })) {
+       *   mentions.push({ asset: 'P-1004A', ...passage.citation });
+       * }
+       * ```
+       */
+      async *allPassages(filter, options) {
+        let cursor = null;
+        while (true) {
+          const page = await this.passages({
+            ...filter !== void 0 ? { filter } : {},
+            cursor,
+            limit: options?.pageSize ?? 200
+          });
+          for (const passage of page.passages) {
+            yield passage;
+          }
+          if (!page.nextCursor || page.passages.length === 0) {
+            return;
+          }
+          cursor = page.nextCursor;
+        }
+      }
       /**
        * Every document in the corpus, oldest first, a page at a time behind the
        * scenes. This is how an app builds its own view of a large corpus (a
@@ -4885,6 +4942,15 @@ var init_source = __esm({
           ...name ? { name } : {}
         });
       }
+      /**
+       * @internal A mapper frame's grant to store the next part of its spill: how
+       * `defineMapper`'s executor hands the platform an object's documents a part
+       * at a time once they outgrow the frame's record. Only a mapper frame's
+       * token is granted one.
+       */
+      async _mapSpillGrant(spillId) {
+        return this._call("map/spill", { slug: this._slug, spillId });
+      }
       /**
        * @internal Content hash of some bytes, matching what the server computes.
        * Exposed for callers that want to check whether they already added a file.
@@ -5066,6 +5132,56 @@ var init_events = __esm({
   }
 });
 
+// src/task/index.ts
+function resolveStepType(name) {
+  const meta = stepMetadata[name];
+  return meta ? meta.stepType : name;
+}
+function mapTools(tools) {
+  return tools.map((t) => {
+    if (typeof t === "object" && "execute" in t) {
+      return {
+        name: t.name,
+        description: t.description,
+        inputSchema: t.inputSchema ?? { type: "object", properties: {} }
+      };
+    }
+    if (typeof t === "object" && "appMethod" in t) {
+      return {
+        appMethod: t.appMethod,
+        ...t.description ? { description: t.description } : {},
+        ...t.defaults ? { defaults: t.defaults } : {}
+      };
+    }
+    const method = typeof t === "string" ? t : t.method;
+    const stepType = resolveStepType(method);
+    const defaults = typeof t === "object" ? t.defaults : void 0;
+    return defaults ? { stepType, defaults } : { stepType };
+  });
+}
+function sleep2(ms) {
+  return new Promise((resolve) => setTimeout(resolve, ms));
+}
+function isDevMode() {
+  return !!(process.env.CALLBACK_TOKEN || getRequestContext()?.callbackToken);
+}
+function logTaskResult(result) {
+  if (!isDevMode()) {
+    return;
+  }
+  const toolSummary = result.toolCalls.map((tc) => `${tc.name} (${tc.durationMs}ms) ${tc.success ? "\u2713" : "\u2717"}`).join(", ");
+  console.log(
+    `[task] ${result.turns} turn${result.turns === 1 ? "" : "s"}` + (toolSummary ? `: ${toolSummary}` : "") + ` | ${result.parsedSuccessfully ? "output OK" : "\u26A0 output not valid JSON"} | cost: ${result.usage.totalBillingCost} | trace: ${result.traceId}`
+  );
+}
+var init_task = __esm({
+  async "src/task/index.ts"() {
+    "use strict";
+    await init_context();
+    init_metadata();
+  }
+});
+
 // src/task/schema.ts
 function describe(value) {
   if (value === void 0) {
@@ -5192,38 +5308,6 @@ function assertSupportedNode(schema, path) {
     assertSupportedNode(schema.items, `${path}[]`);
   }
 }
-function buildExampleFromSchema(schema) {
-  if (schema.const !== void 0) {
-    return schema.const;
-  }
-  if (schema.enum !== void 0 && schema.enum.length > 0) {
-    return schema.enum[0];
-  }
-  const names = Array.isArray(schema.type) ? schema.type : schema.type ? [schema.type] : [];
-  const name = names.find((n) => n !== "null") ?? names[0];
-  switch (name) {
-    case "object": {
-      const example = {};
-      for (const [key, propSchema] of Object.entries(schema.properties ?? {})) {
-        example[key] = buildExampleFromSchema(propSchema);
-      }
-      return example;
-    }
-    case "array":
-      return schema.items ? [buildExampleFromSchema(schema.items)] : [];
-    case "string":
-      return "...";
-    case "number":
-    case "integer":
-      return 0;
-    case "boolean":
-      return true;
-    case "null":
-      return null;
-    default:
-      return null;
-  }
-}
 function stripCodeFences(text) {
   const match = /^```[a-zA-Z]*[ \t]*\r?\n([\s\S]*?)\r?\n?```$/.exec(
     text.trim()
@@ -5252,273 +5336,8 @@ var init_schema = __esm({
   }
 });
 
-// src/task/index.ts
-import { randomUUID as randomUUID2 } from "crypto";
-function resolveStepType(name) {
-  const meta = stepMetadata[name];
-  return meta ? meta.stepType : name;
-}
-function mapTools(tools) {
-  return tools.map((t) => {
-    if (typeof t === "object" && "execute" in t) {
-      return {
-        name: t.name,
-        description: t.description,
-        inputSchema: t.inputSchema ?? { type: "object", properties: {} }
-      };
-    }
-    if (typeof t === "object" && "appMethod" in t) {
-      return {
-        appMethod: t.appMethod,
-        ...t.description ? { description: t.description } : {},
-        ...t.defaults ? { defaults: t.defaults } : {}
-      };
-    }
-    const method = typeof t === "string" ? t : t.method;
-    const stepType = resolveStepType(method);
-    const defaults = typeof t === "object" ? t.defaults : void 0;
-    return defaults ? { stepType, defaults } : { stepType };
-  });
-}
-function buildTaskRequestBody(options) {
-  return {
-    prompt: options.prompt,
-    input: options.input,
-    tools: mapTools(options.tools),
-    // The legacy whole-task route requires structuredOutputExample and
-    // composes its own prompt server-side, so in schema mode we synthesize a
-    // skeleton example from the schema. Validation still happens client-side
-    // after the result comes back (see _runTaskInner).
-    structuredOutputExample: options.outputSchema ? JSON.stringify(buildExampleFromSchema(options.outputSchema)) : typeof options.structuredOutputExample === "string" ? options.structuredOutputExample : JSON.stringify(options.structuredOutputExample),
-    model: options.model,
-    ...options.maxTurns != null && { maxTurns: options.maxTurns },
-    ...options.appId != null && { appId: options.appId },
-    ...options.threadId != null && { threadId: options.threadId }
-  };
-}
-function sleep2(ms) {
-  return new Promise((resolve) => setTimeout(resolve, ms));
-}
-function isDevMode() {
-  return !!(process.env.CALLBACK_TOKEN || getRequestContext()?.callbackToken);
-}
-function logTaskResult(result) {
-  if (!isDevMode()) {
-    return;
-  }
-  const toolSummary = result.toolCalls.map((tc) => `${tc.name} (${tc.durationMs}ms) ${tc.success ? "\u2713" : "\u2717"}`).join(", ");
-  console.log(
-    `[task] ${result.turns} turn${result.turns === 1 ? "" : "s"}` + (toolSummary ? `: ${toolSummary}` : "") + ` | ${result.parsedSuccessfully ? "output OK" : "\u26A0 output not valid JSON"} | cost: ${result.usage.totalBillingCost} | trace: ${result.traceId}`
-  );
-}
-async function runTaskPoll(httpConfig, body) {
-  const traceId = randomUUID2();
-  const { data } = await request(
-    httpConfig,
-    "POST",
-    "/task",
-    body
-  );
-  const pollUrl = `${httpConfig.baseUrl}/developer/v2/task/poll/${data.taskToken}`;
-  let pollDelay = 300;
-  while (true) {
-    await sleep2(pollDelay);
-    pollDelay = Math.min(pollDelay * 1.5, 3e3);
-    const res = await fetch(pollUrl, {
-      headers: { "User-Agent": "@mindstudio-ai/agent" }
-    });
-    if (res.status === 502 || res.status === 503 || res.status === 504) {
-      continue;
-    }
-    if (res.status === 404) {
-      throw new MindStudioError(
-        "Task poll token not found or expired.",
-        "poll_token_expired",
-        404
-      );
-    }
-    if (!res.ok) {
-      const errorBody = await res.json().catch(() => ({}));
-      throw new MindStudioError(
-        errorBody.message ?? errorBody.error ?? `Task poll failed: ${res.status} ${res.statusText}`,
-        errorCodeFrom(errorBody, "poll_error"),
-        res.status,
-        errorBody
-      );
-    }
-    const poll = await res.json();
-    if (poll.status === "pending") {
-      if (isDevMode() && poll.currentTurn != null) {
-        console.log(
-          `[task] running... turn ${poll.currentTurn}/${poll.maxTurns ?? "?"}`
-        );
-      }
-      continue;
-    }
-    if (poll.status === "error") {
-      throw new MindStudioError(
-        poll.error ?? "Task execution failed.",
-        "task_execution_error",
-        500
-      );
-    }
-    const result = {
-      output: poll.output,
-      outputRaw: poll.outputRaw ?? "",
-      parsedSuccessfully: poll.parsedSuccessfully ?? true,
-      turns: poll.turns ?? 0,
-      usage: poll.usage ?? {
-        inputTokens: 0,
-        outputTokens: 0,
-        totalBillingCost: 0
-      },
-      toolCalls: poll.toolCalls ?? [],
-      traceId
-    };
-    logTaskResult(result);
-    return result;
-  }
-}
-async function runTaskStream(httpConfig, body, onEvent) {
-  const traceId = randomUUID2();
-  const url = `${httpConfig.baseUrl}/developer/v2/task`;
-  const res = await fetch(url, {
-    method: "POST",
-    headers: {
-      Authorization: `Bearer ${httpConfig.token}`,
-      "Content-Type": "application/json",
-      "User-Agent": "@mindstudio-ai/agent",
-      Accept: "text/event-stream"
-    },
-    body: JSON.stringify(body)
-  });
-  if (!res.ok) {
-    let message = `${res.status} ${res.statusText}`;
-    let code = "api_error";
-    let details;
-    try {
-      const text = await res.text();
-      try {
-        const errBody = JSON.parse(text);
-        details = errBody;
-        const errMsg = (typeof errBody.error === "string" ? errBody.error : void 0) ?? (typeof errBody.message === "string" ? errBody.message : void 0);
-        if (errMsg) {
-          message = errMsg;
-        }
-        code = errorCodeFrom(errBody, code);
-      } catch {
-        const stripped = text.replace(/<[^>]*>/g, " ").replace(/\s+/g, " ").trim();
-        if (stripped) {
-          message = stripped.slice(0, 200);
-        }
-      }
-    } catch {
-    }
-    throw new MindStudioError(`[task] ${message}`, code, res.status, details);
-  }
-  const reader = res.body.getReader();
-  const decoder = new TextDecoder();
-  let buffer = "";
-  let result = null;
-  while (true) {
-    const { done, value } = await reader.read();
-    if (done) {
-      break;
-    }
-    buffer += decoder.decode(value, { stream: true });
-    const lines = buffer.split("\n");
-    buffer = lines.pop() ?? "";
-    for (const line of lines) {
-      if (!line.startsWith("data: ")) {
-        continue;
-      }
-      try {
-        const event = JSON.parse(line.slice(6));
-        onEvent(event);
-        if (event.type === "error") {
-          throw new MindStudioError(
-            event.error ?? "Task execution failed.",
-            "task_execution_error",
-            500
-          );
-        }
-        if (event.type === "done") {
-          result = {
-            output: event.output,
-            outputRaw: event.outputRaw ?? "",
-            parsedSuccessfully: event.parsedSuccessfully ?? true,
-            turns: event.turns ?? 0,
-            usage: event.usage ?? {
-              inputTokens: 0,
-              outputTokens: 0,
-              totalBillingCost: 0
-            },
-            toolCalls: event.toolCalls ?? [],
-            traceId
-          };
-        }
-      } catch (err) {
-        if (err instanceof MindStudioError) {
-          throw err;
-        }
-      }
-    }
-  }
-  if (buffer.startsWith("data: ")) {
-    try {
-      const event = JSON.parse(buffer.slice(6));
-      onEvent(event);
-      if (event.type === "error") {
-        throw new MindStudioError(
-          event.error ?? "Task execution failed.",
-          "task_execution_error",
-          500
-        );
-      }
-      if (event.type === "done") {
-        result = {
-          output: event.output,
-          outputRaw: event.outputRaw ?? "",
-          parsedSuccessfully: event.parsedSuccessfully ?? true,
-          turns: event.turns ?? 0,
-          usage: event.usage ?? {
-            inputTokens: 0,
-            outputTokens: 0,
-            totalBillingCost: 0
-          },
-          toolCalls: event.toolCalls ?? [],
-          traceId
-        };
-      }
-    } catch (err) {
-      if (err instanceof MindStudioError) {
-        throw err;
-      }
-    }
-  }
-  if (!result) {
-    throw new MindStudioError(
-      "[task] Stream ended without a done event. The task execution may have been interrupted.",
-      "stream_error",
-      500
-    );
-  }
-  logTaskResult(result);
-  return result;
-}
-var init_task = __esm({
-  async "src/task/index.ts"() {
-    "use strict";
-    init_http();
-    init_errors();
-    await init_context();
-    init_metadata();
-    init_schema();
-  }
-});
-
 // src/task/local.ts
-import { randomUUID as randomUUID3 } from "crypto";
+import { randomUUID as randomUUID2 } from "crypto";
 function isPlainObject(value) {
   if (typeof value !== "object" || value === null || Array.isArray(value)) {
     return false;
@@ -5750,7 +5569,7 @@ ${structuredOutputExample}
     Math.max(options.maxTurns || DEFAULT_MAX_TURNS, 1),
     MAX_TURNS_LIMIT
   );
-  const traceId = randomUUID3();
+  const traceId = randomUUID2();
   const messages = [
     { role: "user", content: JSON.stringify(options.input) }
   ];
@@ -5850,13 +5669,6 @@ ${structuredOutputExample}
         traceId
       });
     } catch (err) {
-      if (loopCount === 1 && err instanceof TurnError && err.status === 404 && err.errorCode === "not_found") {
-        throw new MindStudioError(
-          "Task turn endpoint unavailable.",
-          TURN_UNAVAILABLE_CODE,
-          404
-        );
-      }
       return turnFailure(err);
     }
     accumulate(turn);
@@ -6025,7 +5837,7 @@ ${formatValidationErrors(errors)}`
     buildResult(output, finalText, parsedSuccessfully, loopCount + 1)
   );
 }
-var DEFAULT_MAX_TURNS, MAX_TURNS_LIMIT, MAX_SCHEMA_REPAIR_ATTEMPTS, MAX_TOOL_OUTPUT_CHARS, MAX_TURN_ATTEMPTS, INITIAL_BACKOFF_MS, STALL_TIMEOUT_MS, TURN_UNAVAILABLE_CODE, UNSAFE_KEYS, TurnError;
+var DEFAULT_MAX_TURNS, MAX_TURNS_LIMIT, MAX_SCHEMA_REPAIR_ATTEMPTS, MAX_TOOL_OUTPUT_CHARS, MAX_TURN_ATTEMPTS, INITIAL_BACKOFF_MS, STALL_TIMEOUT_MS, UNSAFE_KEYS, TurnError;
 var init_local = __esm({
   async "src/task/local.ts"() {
     "use strict";
@@ -6039,7 +5851,6 @@ var init_local = __esm({
     MAX_TURN_ATTEMPTS = 5;
     INITIAL_BACKOFF_MS = 1e3;
     STALL_TIMEOUT_MS = 3e5;
-    TURN_UNAVAILABLE_CODE = "task_turn_unavailable";
     UNSAFE_KEYS = /* @__PURE__ */ new Set(["__proto__", "constructor", "prototype"]);
     TurnError = class extends Error {
       constructor(message, retryable, phase, status, errorCode) {
@@ -6587,9 +6398,7 @@ var init_client = __esm({
     init_analytics();
     init_email();
     init_events();
-    await init_task();
     await init_local();
-    init_schema();
     init_steps();
     init_metadata();
     DEFAULT_BASE_URL = "https://v1.mindstudio-api.com";
@@ -7182,108 +6991,75 @@ var init_client = __esm({
       }
       async _runTaskInner(options) {
         const httpConfig = this._currentHttpConfig;
-        try {
-          return await runTaskLocal(
-            {
-              httpConfig,
-              executeStepTool: async (stepType, input) => {
-                try {
-                  const result = await this.executeStep(stepType, input, {
-                    requestSource: "v2-task"
-                  });
-                  const output = {};
-                  let billingCost = 0;
-                  for (const [key, value] of Object.entries(
-                    result
-                  )) {
-                    if (key === "$billingCost" && typeof value === "number") {
-                      billingCost = value;
-                    }
-                    if (!key.startsWith("$")) {
-                      output[key] = value;
-                    }
+        return runTaskLocal(
+          {
+            httpConfig,
+            executeStepTool: async (stepType, input) => {
+              try {
+                const result = await this.executeStep(stepType, input, {
+                  requestSource: "v2-task"
+                });
+                const output = {};
+                let billingCost = 0;
+                for (const [key, value] of Object.entries(
+                  result
+                )) {
+                  if (key === "$billingCost" && typeof value === "number") {
+                    billingCost = value;
                   }
-                  return { output, billingCost, isError: false };
-                } catch (err) {
-                  return {
-                    output: {
-                      error: err instanceof Error ? err.message : "Step execution failed"
-                    },
-                    billingCost: 0,
-                    isError: true
-                  };
-                }
-              },
-              executeMethodTool: async (methodId, input, call) => {
-                try {
-                  const { data } = await request(
-                    { ...httpConfig, maxRetries: 0 },
-                    "POST",
-                    "/task/invoke-method",
-                    {
-                      methodId,
-                      input,
-                      traceId: call.traceId,
-                      toolCallId: call.toolCallId
-                    }
-                  );
-                  if (data.error) {
-                    return {
-                      output: { error: data.error },
-                      billingCost: 0,
-                      isError: true
-                    };
+                  if (!key.startsWith("$")) {
+                    output[key] = value;
                   }
-                  return {
-                    output: data.output ?? null,
-                    billingCost: 0,
-                    isError: false
-                  };
-                } catch (err) {
-                  return {
-                    output: {
-                      error: err instanceof Error ? err.message : "Method execution failed"
-                    },
-                    billingCost: 0,
-                    isError: true
-                  };
                 }
+                return { output, billingCost, isError: false };
+              } catch (err) {
+                return {
+                  output: {
+                    error: err instanceof Error ? err.message : "Step execution failed"
+                  },
+                  billingCost: 0,
+                  isError: true
+                };
               }
             },
-            options
-          );
-        } catch (err) {
-          if (err instanceof MindStudioError && err.code === TURN_UNAVAILABLE_CODE) {
-            if (options.tools.some((t) => typeof t === "object" && "execute" in t)) {
-              throw new MindStudioError(
-                "[task] This server does not support the per-turn task endpoint required for function tools.",
-                "task_function_tools_unsupported",
-                404
-              );
-            }
-            const body = buildTaskRequestBody(options);
-            const result = options.onEvent ? await runTaskStream(httpConfig, body, options.onEvent) : await runTaskPoll(httpConfig, body);
-            if (options.outputSchema) {
-              const errors = result.parsedSuccessfully ? validateAgainstSchema(result.output, options.outputSchema) : [{ path: "$", message: "output was not valid JSON" }];
-              if (errors.length > 0) {
-                throw new MindStudioError(
-                  "[task] Output did not conform to outputSchema (legacy task route, no repair turns).",
-                  "task_output_schema_mismatch",
-                  422,
+            executeMethodTool: async (methodId, input, call) => {
+              try {
+                const { data } = await request(
+                  { ...httpConfig, maxRetries: 0 },
+                  "POST",
+                  "/task/invoke-method",
                   {
-                    outputRaw: result.outputRaw,
-                    errors,
-                    turns: result.turns,
-                    usage: result.usage,
-                    toolCalls: result.toolCalls
+                    methodId,
+                    input,
+                    traceId: call.traceId,
+                    toolCallId: call.toolCallId
                   }
                 );
+                if (data.error) {
+                  return {
+                    output: { error: data.error },
+                    billingCost: 0,
+                    isError: true
+                  };
+                }
+                return {
+                  output: data.output ?? null,
+                  billingCost: 0,
+                  isError: false
+                };
+              } catch (err) {
+                return {
+                  output: {
+                    error: err instanceof Error ? err.message : "Method execution failed"
+                  },
+                  billingCost: 0,
+                  isError: true
+                };
               }
             }
-            return result;
-          }
-          throw err;
-        }
+          },
+          options
+        );
       }
       /**
        * Get the authenticated user's identity and organization info.
@@ -9104,7 +8880,7 @@ var llmsContent;
 var init_llms_content = __esm({
   "src/generated/llms-content.ts"() {
     "use strict";
-    llmsContent = '# @mindstudio-ai/agent\n\nTypeScript SDK, CLI, and MCP server for MindStudio. One API key gives you access to 200+ AI models (OpenAI, Anthropic, Google, Meta, xAI, DeepSeek, etc.) and 1,000+ actions including 850+ connector actions across third-party services from the open-source MindStudio Connector Registry (https://github.com/mindstudio-ai/mscr). No separate provider API keys required.\n\nThis file is the complete API reference. No other documentation is needed to use the SDK.\n\n## Recommended workflow\n\nThere are 150+ actions available. Do NOT try to read or load them all at once. Follow this discovery flow:\n\n1. **Identify yourself** \u2014 Call `changeName` to set your display name (use your name or whatever your user calls you). If you have a profile picture, pass its public image URL to `changeProfilePicture`. This helps users identify your requests in their logs.\n2. **Ask** \u2014 Use `mindstudio ask "your question"` (CLI) or the `ask` MCP tool for SDK guidance. It knows every action, model, and connector and returns working TypeScript code with real model IDs and config options. Examples: `mindstudio ask "generate an image with FLUX"`, `mindstudio ask "what models support vision?"`, `mindstudio ask "how do I send a Slack message?"`.\n3. **Browse** \u2014 For manual discovery, call `listActions` (MCP tool) or `mindstudio list-actions --summary` (CLI) to get a compact `{ action: description }` map of everything available (~3k tokens). Call `mindstudio info <action>` (CLI) for parameter details.\n4. **Call it** \u2014 Invoke the action with the required parameters. All actions share the same calling convention (see below).\n\nFor specific use cases:\n\n- **OAuth third-party integrations** (Slack, Google, HubSpot, etc.): These are optional OAuth connectors from the MindStudio Connector Registry \u2014 for most tasks, use actions directly instead. If you need a third-party integration: call `listConnectors()` to browse services \u2192 `getConnectorAction(serviceId, actionId)` for input fields \u2192 execute via `runFromConnectorRegistry`. Requires an OAuth connection set up in MindStudio first \u2014 call `listConnections()` to check available connections.\n- **Pre-built agents**: Call `listAgents()` to see what\'s available \u2192 `runAgent({ appId })` to execute one. **Important:** Not all agents are configured for API use. Do not try to run an agent just because it appears in the list \u2014 only run agents the user specifically asks you to run.\n- **Model selection**: Call `listModelsSummary()` or `listModelsSummaryByType("llm_chat")` to browse models, then pass the model ID as `modelOverride.model` to actions like `generateText`. Use the summary endpoints (not `listModels`) to keep token usage low.\n- **Cost estimation**: AI-powered actions (text generation, image generation, video, audio, etc.) cost money. Call `estimateStepCost(stepType, stepInput)` before running these and confirm with the user before proceeding \u2014 unless they\'ve explicitly given permission to go ahead. Non-AI actions (data lookups, OAuth connectors, etc.) are generally free.\n- **Task agents**: For multi-step tasks that need autonomous tool use (research, content creation, data enrichment), use `runTask()`. Provide a prompt, input, tools, an output contract, and a tool-use model id. The SDK runs a tool-use loop \u2014 calling the model, executing tools, feeding results back \u2014 until structured output is produced. Tools are SDK action names, your own app methods written as `{ appMethod: "save-vendor", description: "when to use it" }`, and/or inline functions written as `{ name, description, inputSchema, execute }` \u2014 the function runs in your process when the model calls it, and a thrown error goes back to the model as tool output to work around. App methods run as the user who invoked the method that started the task, with their roles, so the agent can read and write app state on their behalf. Prefer `outputSchema` for the output contract: plain JSON Schema (`type`/`properties`/`required`/`enum`/`items`; nullable via type arrays like `["string","null"]` \u2014 no `oneOf`/`$ref`/`nullable`). Output is validated every turn with automatic repair, `result.output` is fully typed from the schema (no generic argument, no `parsedSuccessfully` check), and the call throws `MindStudioError` code `task_output_schema_mismatch` if the agent cannot produce conforming output. Example: `runTask({ prompt: "Research this restaurant and save it", input: { name: "Tartine" }, tools: ["searchGoogle", "fetchUrl", { appMethod: "saveRestaurant", description: "Persist the researched restaurant." }], outputSchema: { type: "object", properties: { name: { type: "string" }, url: { type: ["string", "null"] } }, required: ["name"] }, model: yourModelId })`. The legacy `structuredOutputExample` option (an example object the model imitates) still works but is unvalidated \u2014 check `result.parsedSuccessfully` before using its output. Pass a current tool-use model id (call `listModels` or ask). Don\'t hardcode one.\n\n## Install\n\nStandalone binary (CLI/MCP, no dependencies):\n```bash\ncurl -fsSL https://msagent.ai/install.sh | bash\n```\n\nnpm (SDK + CLI):\n```bash\nnpm install @mindstudio-ai/agent\n```\n\nRequires Node.js >= 18.\n\n## CLI\n\nThe package includes a CLI for executing steps from the command line or scripts:\n\n```bash\n# Execute with named flags (kebab-case)\nmindstudio generate-image --prompt "A mountain landscape"\n\n# Execute with JSON input (JSON5-tolerant)\nmindstudio generate-image \'{prompt: "A mountain landscape"}\'\n\n# Extract a single output field\nmindstudio generate-image --prompt "A sunset" --output-key imageUrl\n\n# List all methods (compact JSON \u2014 best for LLM discovery)\nmindstudio list --summary\n\n# List all methods (human-readable table)\nmindstudio list\n\n# Show method details (params, types, output)\nmindstudio info generate-image\n\n# Run via npx without installing\nnpx @mindstudio-ai/agent generate-text --message "Hello"\n```\n\nAuth: run `mindstudio login`, set `MINDSTUDIO_API_KEY` env var, or pass `--api-key <key>`.\nMethod names are kebab-case on the CLI (camelCase also accepted). Flags are kebab-case (`--video-url` for `videoUrl`).\nUse `--output-key <key>` to extract a single field, `--no-meta` to strip $-prefixed metadata.\n\n### Authentication\n\n```bash\n# Interactive login (opens browser, saves key to ~/.mindstudio/config.json)\nmindstudio login\n\n# Check current auth status\nmindstudio whoami\n\n# Clear stored credentials\nmindstudio logout\n```\n\nAuth resolution order: `--api-key` flag > `MINDSTUDIO_API_KEY` env > `~/.mindstudio/config.json` > `CALLBACK_TOKEN` env.\n\n## MCP server\n\nThe package includes an MCP server exposing all methods as tools. Start by calling the `listSteps` tool to discover available methods.\n\n```bash\nmindstudio mcp\n```\n\nMCP client config (standalone binary \u2014 recommended):\n```json\n{\n  "mcpServers": {\n    "mindstudio": {\n      "command": "mindstudio",\n      "args": ["mcp"],\n      "env": { "MINDSTUDIO_API_KEY": "your-api-key" }\n    }\n  }\n}\n```\n\n## Setup\n\nInside MindStudio apps (auth is automatic):\n```typescript\nimport { mindstudio, db, auth, Roles, stream } from \'@mindstudio-ai/agent\';\n\nconst { imageUrl } = await mindstudio.generateImage({ prompt: \'a sunset\' });\n```\n\nExternal usage (API key required):\n```typescript\nimport { MindStudioAgent } from \'@mindstudio-ai/agent\';\nconst agent = new MindStudioAgent({ apiKey: \'your-key\' });\n```\n\nYour MindStudio API key authenticates all requests. MindStudio routes to the correct AI provider (OpenAI, Google, Anthropic, etc.) server-side \u2014 you do NOT need separate provider API keys.\n\n## Models\n\nDirect access to 200+ AI models from every major provider \u2014 all through a single API key, billed at cost with no markups.\n\nUse `listModels()` or `listModelsByType()` for full model details, or `listModelsSummary()` / `listModelsSummaryByType()` for a lightweight list (id, name, type, tags) suitable for LLM context windows. Pass a model ID to `modelOverride.model` in methods like `generateText` to select a specific model:\n\n```typescript\nconst { models } = await mindstudio.listModelsByType(\'llm_chat\');\nconst model = models.find(m => m.name.includes("Gemini"));\n\nconst { content } = await mindstudio.generateText({\n  message: \'Hello\',\n  modelOverride: {\n    model: model.id,\n    temperature: 0.7,\n    maxResponseTokens: 16000,\n  },\n});\n```\n\n## Calling convention\n\nEvery method has the signature:\n```typescript\nmindstudio.methodName(input: InputType, options?: { appId?, threadId?, onLog? }): Promise<OutputType & StepExecutionMeta>\n```\n\nThe first argument is the step-specific input object. The optional second argument controls thread/app context and debug logging.\n\n**Debug logging**: Pass `onLog` in the options to get real-time debug logs during execution (SSE streaming). Works on ALL step methods:\n```typescript\nawait mindstudio.generateImage({ prompt: \'...\' }, {\n  onLog: (event) => console.log(`[${event.tag}] ${event.value}`),\n});\n// event: { value: string, tag: string, ts: number }\n```\n\n**`stream()` \u2014 push real-time updates to the frontend:**\n```typescript\nimport { mindstudio, stream } from \'@mindstudio-ai/agent\';\n\n// String \u2192 \'token\' event (frontend receives via onToken, accumulated text)\nawait stream(\'Processing...\');\n\n// Object \u2192 \'data\' event (frontend receives via onStreamData, NOT onToken)\nawait stream({ status: \'generating\', progress: 50 });\n```\nThese are different event types \u2014 do not mix them up. `stream()` is silently ignored when no SSE connection is active (safe to include unconditionally). For long-running steps, combine `onLog` with `stream()` to push progress to the frontend.\n\n**Results are returned flat** \u2014 output fields are spread at the top level alongside metadata:\n\n```typescript\nconst { content } = await mindstudio.generateText({ message: \'Hello\' });\n\n// Full result shape for any method:\nconst result = await mindstudio.generateText({ message: `Hello` });\nresult.content;              // step-specific output field\nresult.$appId;               // string \u2014 app ID for this execution\nresult.$threadId;            // string \u2014 thread ID for this execution\nresult.$rateLimitRemaining;  // number | undefined \u2014 API calls remaining in rate limit window\nresult.$billingCost;         // number | undefined \u2014 cost in nanodollars (1e-9 USD; divide by 1e9 for dollars, e.g. 40000000 = $0.04)\nresult.$billingEvents;       // object[] | undefined \u2014 itemized billing events\n```\n\n## Thread persistence\n\nPass `$appId`/`$threadId` from a previous result to maintain conversation state, variable state, or other context across calls:\n\n```typescript\nconst r1 = await mindstudio.generateText({ message: \'My name is Alice\' });\nconst r2 = await mindstudio.generateText(\n  { message: \'What is my name?\' },\n  { threadId: r1.$threadId, appId: r1.$appId },\n);\n// r2.content => "Your name is Alice"\n```\n\n## Error handling\n\nAll errors throw `MindStudioError`:\n```typescript\nimport { MindStudioError } from \'@mindstudio-ai/agent\';\n\ntry {\n  await mindstudio.generateImage({ prompt: \'...\' });\n} catch (err) {\n  if (err instanceof MindStudioError) {\n    err.message; // Human-readable error message\n    err.code;    // Machine-readable code: "invalid_step_config", "api_error", "rate_limit_exceeded", "output_fetch_error"\n    err.status;  // HTTP status code (400, 401, 429, etc.)\n    err.details; // Raw error body from the API\n  }\n}\n```\n\n429 rate limit errors are retried automatically (configurable via `maxRetries`).\n\n## Low-level access\n\nFor action types not covered by generated methods:\n```typescript\nconst result = await mindstudio.executeStep(\'stepType\', { ...params });\n```\n\n## Batch execution\n\nExecute multiple steps in parallel in a single request. Maximum 50 steps per batch.\nIndividual step failures do not affect other steps \u2014 partial success is possible.\n\n```typescript\nconst result = await mindstudio.executeStepBatch([\n  { stepType: \'generateImage\', step: { prompt: \'a sunset\' } },\n  { stepType: \'textToSpeech\', step: { text: \'hello world\' } },\n], { appId?, threadId? });\n\n// Result:\nresult.results;          // BatchStepResult[] \u2014 same order as input\nresult.results[0].stepType;  // string\nresult.results[0].output;    // object | undefined (step output on success)\nresult.results[0].error;     // string | undefined (error message on failure)\nresult.results[0].billingCost; // number | undefined (cost in nanodollars on success)\nresult.totalBillingCost;  // number | undefined \u2014 nanodollars (divide by 1e9 for USD)\nresult.appId;            // string\nresult.threadId;         // string\n```\n\nCLI:\n```bash\nmindstudio batch \'[{"stepType":"generateImage","step":{"prompt":"a cat"}}]\'\ncat steps.json | mindstudio batch\n```\n\n## Methods\n\nAll methods below are called on a `MindStudioAgent` instance (`agent.methodName(...)`).\nInput shows the first argument object. Output shows the fields available on the returned result.\n\n### General\n\n#### addSubtitlesToVideo\nAutomatically add subtitles to a video\n- Can control style of text and animation\n- Input: `{ videoUrl: string, language: string, fontName: string, fontSize: number, fontWeight: "normal" | "bold" | "black", fontColor: "white" | "black" | "red" | "green" | "blue" | "yellow" | "orange" | "purple" | "pink" | "brown" | "gray" | "cyan" | "magenta", highlightColor: "white" | "black" | "red" | "green" | "blue" | "yellow" | "orange" | "purple" | "pink" | "brown" | "gray" | "cyan" | "magenta", strokeWidth: number, strokeColor: "black" | "white" | "red" | "green" | "blue" | "yellow" | "orange" | "purple" | "pink" | "brown" | "gray" | "cyan" | "magenta", backgroundColor: "black" | "white" | "red" | "green" | "blue" | "yellow" | "orange" | "purple" | "pink" | "brown" | "gray" | "cyan" | "magenta" | "none", backgroundOpacity: number, position: "top" | "center" | "bottom", yOffset: number, wordsPerSubtitle: number, enableAnimation: boolean, intermediateAsset?: boolean }`\n- Output: `{ videoUrl: string }`\n\n#### analyzeImage\nAnalyze an image using a vision model based on a text prompt.\n- Uses the configured vision model to generate a text analysis of the image.\n- The prompt should describe what to look for or extract from the image.\n- Pass imageUrl for a single image, or imageUrls for multiple images analyzed together in one request.\n- Most vision models (OpenAI, Grok, Gemini) accept multiple images in one request. Ideogram describe is single-image only.\n- Input: `{ prompt: string, imageUrl?: string, imageUrls?: string[], visionModelOverride?: { model: string, config?: object } | { model: string, temperature: number, maxResponseTokens: number, ignorePreamble?: boolean, userMessagePreprocessor?: { dataSource?: string, messageTemplate?: string, maxResults?: number, enabled?: boolean, shouldInherit?: boolean }, preamble?: string, multiModelEnabled?: boolean, editResponseEnabled?: boolean, promptCaching?: boolean, config?: object } }`\n- Output: `{ analysis: string }`\n\n#### analyzeVideo\nAnalyze a video using a video analysis model based on a text prompt.\n- Uses the configured video analysis model to generate a text analysis of the video.\n- The prompt should describe what to look for or extract from the video.\n- Input: `{ prompt: string, videoUrl: string, videoAnalysisModelOverride?: { model: string, config?: object } | { model: string, temperature: number, maxResponseTokens: number, ignorePreamble?: boolean, userMessagePreprocessor?: { dataSource?: string, messageTemplate?: string, maxResults?: number, enabled?: boolean, shouldInherit?: boolean }, preamble?: string, multiModelEnabled?: boolean, editResponseEnabled?: boolean, promptCaching?: boolean, config?: object } }`\n- Output: `{ analysis: string }`\n\n#### captureThumbnail\nCapture a thumbnail from a video at a specified timestamp\n- Input: `{ videoUrl: string, at: number | string }`\n- Output: `{ thumbnailUrl: string }`\n\n#### checkAppRole\nCheck whether the current user has a specific app role and branch accordingly.\n- Checks if the current user has been assigned a specific role in this app.\n- If the user has the role, transitions to the "has role" path.\n- If the user does not have the role, transitions to the "no role" path, or errors if no path is configured.\n- Role names are defined by the app creator and assigned to users via the app roles system.\n- The roleName field supports {{variables}} for dynamic role checks.\n- Input: `{ roleName: string, hasRoleStepId?: string, hasRoleWorkflowId?: string, noRoleStepId?: string, noRoleWorkflowId?: string }`\n- Output: `{ hasRole: boolean, userRoles: string[] }`\n\n#### convertPdfToImages\nConvert each page of a PDF document into a PNG image.\n- Each page is converted to a separate PNG.\n- The page images go to the public CDN by default, or into one of the app\'s own file stores when the call passes `store` \u2014 a private store keeps them private.\n- The source PDF is never published: one of the app\'s own files is read where it sits, and any other URL is staged privately for the renderer and discarded afterwards.\n- Returns an array of image URLs, one per page.\n- Input: `{ pdfUrl: string }`\n- Output: `{ imageUrls: string[] }`\n\n#### createDataSource\nLegacy v1 data sources. v2 apps should use `dataSources.defineDataSource()` from @mindstudio-ai/agent instead, which owns parsing, chunking, embedding and citations. Create a new empty vector data source for the current app.\n- Creates a new data source (vector database) associated with the current app version.\n- The data source is created empty \u2014 use the "Upload Data Source Document" block to add documents.\n- Returns the new data source ID which can be used in subsequent blocks.\n- Input: `{ name: string }`\n- Output: `unknown`\n\n#### createGmailDraft\nCreate a draft email in the connected Gmail account.\n- Requires a Google OAuth connection with Gmail compose scope.\n- The draft appears in the user\'s Gmail Drafts folder but is not sent.\n- messageType controls the body format: "plain" for plain text, "html" for raw HTML, "markdown" for auto-converted markdown.\n- Input: `{ to: string, subject: string, message: string, connectionId?: string, messageType: "plain" | "html" | "markdown" }`\n- Output: `{ draftId: string }`\n\n#### decomposeImage\nSplit an image into a base image and transparent layers.\n- Uses Seedream 5.0 Flash layer decomposition. There is no model picker.\n- An optional prompt can name the objects or groupings to separate.\n- The output is an array of hosted image URLs. PNG output keeps layer transparency.\n- Input: `{ image: string, prompt?: string, intermediateAsset?: boolean }`\n- Output: `{ imageUrls: string[] }`\n\n#### deleteDataSource\nLegacy v1 data sources. v2 apps should use `dataSources.defineDataSource()` from @mindstudio-ai/agent instead, which owns parsing, chunking, embedding and citations. Delete a vector data source from the current app.\n- Soft-deletes a data source (vector database) by marking it as deleted.\n- The Milvus partition is cleaned up asynchronously by a background cron job.\n- The data source must belong to the current app version.\n- Input: `{ dataSourceId: string }`\n- Output: `unknown`\n\n#### deleteDataSourceDocument\nLegacy v1 data sources. v2 apps should use `dataSources.defineDataSource()` from @mindstudio-ai/agent instead, which owns parsing, chunking, embedding and citations. Delete a single document from a data source.\n- Soft-deletes a document by marking it as deleted.\n- Requires both the data source ID and document ID.\n- After deletion, reloads vectors into Milvus so the data source reflects the change immediately.\n- Input: `{ dataSourceId: string, documentId: string }`\n- Output: `unknown`\n\n#### detectChanges\nDetect changes between runs by comparing current input against previously stored state. Routes execution based on whether a change occurred.\n- Persists state across runs using a global variable keyed to the step ID.\n- Two modes: "comparison" (default) uses strict string inequality; "ai" uses an LLM to determine if a meaningful change occurred.\n- First run always treats the value as "changed" since there is no previous state.\n- Each mode supports transitions to different steps/workflows for the "changed" and "unchanged" paths.\n- AI mode bills normally for the LLM call.\n- Input: `{ mode: "ai" | "comparison", input: string, prompt?: string, modelOverride?: { model: string, temperature: number, maxResponseTokens: number, ignorePreamble?: boolean, userMessagePreprocessor?: { dataSource?: string, messageTemplate?: string, maxResults?: number, enabled?: boolean, shouldInherit?: boolean }, preamble?: string, multiModelEnabled?: boolean, editResponseEnabled?: boolean, promptCaching?: boolean, config?: object }, previousValueVariable?: string, changedStepId?: string, changedWorkflowId?: string, unchangedStepId?: string, unchangedWorkflowId?: string }`\n- Output: `{ hasChanged: boolean, currentValue: string, previousValue: string, isFirstRun: boolean }`\n\n#### detectPII\nScan text for personally identifiable information using Microsoft Presidio.\n- In workflow mode, transitions to detectedStepId if PII is found, notDetectedStepId otherwise.\n- In direct execution, returns the detection results without transitioning.\n- If entities is empty, returns immediately with no detections.\n- Input: `{ input: string, language: string, entities: string[], detectedStepId?: string, notDetectedStepId?: string, outputLogVariable?: string | null }`\n- Output: `{ detected: boolean, detections: { entity_type: string, start: number, end: number, score: number }[] }`\n\n#### downloadVideo\nDownload a video file\n- Works with YouTube, TikTok, etc., by using ytdlp behind the scenes\n- Can save as mp4 or mp3\n- Input: `{ videoUrl: string, format: "mp4" | "mp3" }`\n- Output: `{ videoUrl: string }`\n\n#### enhanceImageGenerationPrompt\nGenerate or enhance an image generation prompt using a language model. Optionally generates a negative prompt.\n- Rewrites the user\'s prompt with added detail about style, lighting, colors, and composition.\n- Also useful for initial generation, it doesn\'t always need to be enhancing an existing prompt\n- When includeNegativePrompt is true, a second model call generates a negative prompt.\n- Input: `{ initialPrompt: string, includeNegativePrompt: boolean, negativePromptDestinationVariableName?: string, systemPrompt: string, modelOverride?: unknown }`\n- Output: `{ prompt: string, negativePrompt?: string }`\n\n#### enhanceVideoGenerationPrompt\nGenerate or enhance a video generation prompt using a language model. Optionally generates a negative prompt.\n- Rewrites the user\'s prompt with added detail about style, camera movement, lighting, and composition.\n- Also useful for initial generation, it doesn\'t always need to be enhancing an existing prompt\n- When includeNegativePrompt is true, a second model call generates a negative prompt.\n- Input: `{ initialPrompt: string, includeNegativePrompt: boolean, negativePromptDestinationVariableName?: string, systemPrompt: string, modelOverride?: unknown }`\n- Output: `{ prompt: string, negativePrompt?: string }`\n\n#### extractAudioFromVideo\nExtract audio MP3 from a video file\n- Input: `{ videoUrl: string }`\n- Output: `{ audioUrl: string }`\n\n#### extractText\nDownload a file from a URL and extract its text content. Supports PDFs (including scanned/image-based PDFs via OCR), plain text files, and other document formats.\n- Best suited for PDFs and raw text/document files. For web pages, use the scrapeUrl step instead.\n- Handles both text-layer PDFs and image-based/scanned PDFs (e.g. Figma/Canva exports, scanned documents). Image-based PDFs are processed with OCR automatically \u2014 there is no need to convert PDF pages to images first.\n- Accepts a single URL, a comma-separated list of URLs, or a JSON array of URLs.\n- Nothing is published. One of the app\'s own files \u2014 including a private one, or a share link \u2014 is read where it sits; any other URL is downloaded and staged privately for the extraction model, then discarded when extraction finishes.\n- Optionally set `model` to a specific document-extraction model (`mistral-ocr-latest`, `llamaparse`, `google-document-ai`); omit to use the platform default.\n- Maximum file size is 50MB per URL.\n- Input: `{ url: string | string[], model?: string }`\n- Output: `{ text: string | string[] }`\n\n#### fetchDataSourceDocument\nLegacy v1 data sources. v2 apps should use `dataSources.defineDataSource()` from @mindstudio-ai/agent instead, which owns parsing, chunking, embedding and citations. Fetch the full extracted text contents of a document in a data source.\n- Loads a document by ID and returns its full extracted text content.\n- The document must have been successfully processed (status "done").\n- Also returns document metadata (name, summary, word count).\n- Input: `{ dataSourceId: string, documentId: string }`\n- Output: `unknown`\n\n#### fetchSlackChannelHistory\nFetch recent message history from a Slack channel.\n- The user is responsible for connecting their Slack workspace and selecting the channel\n- Input: `{ connectionId?: string, channelId: string, limit?: number, startDate?: string, endDate?: string, includeImages?: boolean, includeRawMessage?: boolean }`\n- Output: `{ messages: { from: string, content: string, timestamp?: string, images?: string[], rawMessage?: { agent_session?: { agent_bot_user_ids?: string[], agent_statuses?: unknown[], date_status_processing_expire?: number, status?: string, title?: string }, app_id?: string, assistant_app_thread?: { first_user_thread_reply?: string, title?: string, title_blocks?: unknown[] }, attachments?: { actions?: unknown[], app_id?: string, app_unfurl_url?: string, author_icon?: string, author_id?: string, author_link?: string, author_name?: string, author_subname?: string, blocks?: unknown[], bot_id?: string, bot_team_id?: string, callback_id?: string, channel_id?: string, channel_name?: string, channel_team?: string, color?: string, fallback?: string, fields?: unknown[], file_id?: string, filename?: string, files?: unknown[], footer?: string, footer_icon?: string, from_url?: string, hide_border?: boolean, hide_color?: boolean, id?: number, image_bytes?: number, image_height?: number, image_url?: string, image_width?: number, indent?: boolean, is_app_unfurl?: boolean, is_file_attachment?: boolean, is_msg_unfurl?: boolean, is_reply_unfurl?: boolean, is_thread_root_unfurl?: boolean, list?: unknown, list_record?: unknown, list_record_id?: string, list_records?: unknown[], list_schema?: unknown[], list_view?: unknown, list_view_id?: string, message_blocks?: unknown[], metadata?: unknown, mimetype?: string, mrkdwn_in?: string[], msg_subtype?: string, original_url?: string, pretext?: string, preview?: unknown, service_icon?: string, service_name?: string, service_url?: string, size?: number, slack_file_id?: string, text?: string, thumb_height?: number, thumb_url?: string, thumb_width?: number, title?: string, title_link?: string, ts?: string, url?: string, video_html?: string, video_html_height?: number, video_html_width?: number, video_url?: string }[], blocks?: { accessory?: unknown, alt_text?: string, api_decoration_available?: boolean, app_collaborators?: string[], app_id?: string, author_name?: string, block_id?: string, bot_user_id?: string, button_label?: string, call?: unknown, call_id?: string, description?: unknown, developer_trace_id?: string, dispatch_action?: boolean, element?: unknown, elements?: unknown[], expand?: boolean, external_id?: string, fallback?: string, fields?: unknown[], file?: unknown, file_id?: string, function_trigger_id?: string, hint?: unknown, image_bytes?: number, image_height?: number, image_url?: string, image_width?: number, is_animated?: boolean, is_workflow_app?: boolean, label?: unknown, optional?: boolean, owning_team_id?: string, provider_icon_url?: string, provider_name?: string, sales_home_workflow_app_type?: number, share_url?: string, slack_file?: unknown, source?: string, text?: unknown, thumbnail_url?: string, title?: unknown, title_url?: string, trigger_subtype?: string, trigger_type?: string, type?: unknown, url?: string, video_url?: string, workflow_id?: string }[], bot_id?: string, bot_profile?: { app_id?: string, deleted?: boolean, icons?: unknown, id?: string, name?: string, team_id?: string, updated?: number, user_id?: string }, client_msg_id?: string, display_as_bot?: boolean, edited?: { ts?: string, user?: string }, files?: { access?: string, alt_txt?: string, app_id?: string, app_name?: string, app_provides_file_work_objects?: boolean, attachments?: unknown[], blocks?: unknown[], bot_id?: string, can_toggle_canvas_lock?: boolean, canvas_creator_id?: string, canvas_printing_enabled?: boolean, canvas_readtime?: number, canvas_template_mode?: string, cc?: unknown[], channel_actions_count?: number, channel_actions_ts?: string, channels?: string[], comments_count?: number, converted_pdf?: string, created?: number, deanimate?: string, deanimate_gif?: string, display_as_bot?: boolean, dm_mpdm_users_with_file_access?: unknown[], duration_ms?: number, edit_link?: string, edit_timestamp?: number, editable?: boolean, editor?: string, editors?: string[], external_id?: string, external_type?: string, external_url?: string, favorites?: unknown[], file_access?: string, filetype?: string, from?: unknown[], groups?: string[], has_more?: boolean, has_more_shares?: boolean, has_rich_preview?: boolean, headers?: unknown, hls?: string, hls_embed?: string, id?: string, image_exif_rotation?: number, ims?: string[], initial_comment?: unknown, inline_attachment_count?: number, is_ai_suggested?: boolean, is_channel_space?: boolean, is_external?: boolean, is_modified_by_ai?: boolean, is_public?: boolean, is_restricted_sharing_enabled?: boolean, is_starred?: boolean, last_editor?: string, last_read?: number, lines?: number, lines_more?: number, linked_channel_id?: string, list_csv_download_url?: string, list_limits?: unknown, list_metadata?: unknown, media_display_type?: string, media_progress?: unknown, mimetype?: string, mode?: string, mp4?: string, mp4_low?: string, name?: string, non_owner_editable?: boolean, num_stars?: number, org_or_workspace_access?: string, original_attachment_count?: number, original_h?: string, original_w?: string, permalink?: string, permalink_public?: string, pinned_to?: string[], pjpeg?: string, plain_text?: string, pretty_type?: string, preview?: string, preview_highlight?: string, preview_is_truncated?: boolean, preview_plain_text?: string, private_channels_with_file_access_count?: number, private_file_with_access_count?: number, public_url_shared?: boolean, quip_thread_id?: string, reactions?: unknown[], saved?: unknown, sent_to_self?: boolean, shares?: unknown, show_badge?: boolean, simplified_html?: string, size?: number, skipped_shares?: boolean, source_team?: string, subject?: string, subtype?: string, team_pref_version_history_enabled?: boolean, teams_shared_with?: unknown[], template_conversion_ts?: number, template_description?: string, template_icon?: string, template_name?: string, template_title?: string, thumb_1024?: string, thumb_1024_gif?: string, thumb_1024_h?: string, thumb_1024_w?: string, thumb_160?: string, thumb_160_gif?: string, thumb_160_h?: string, thumb_160_w?: string, thumb_360?: string, thumb_360_gif?: string, thumb_360_h?: string, thumb_360_w?: string, thumb_480?: string, thumb_480_gif?: string, thumb_480_h?: string, thumb_480_w?: string, thumb_64?: string, thumb_64_gif?: string, thumb_64_h?: string, thumb_64_w?: string, thumb_720?: string, thumb_720_gif?: string, thumb_720_h?: string, thumb_720_w?: string, thumb_80?: string, thumb_800?: string, thumb_800_gif?: string, thumb_800_h?: string, thumb_800_w?: string, thumb_80_gif?: string, thumb_80_h?: string, thumb_80_w?: string, thumb_960?: string, thumb_960_gif?: string, thumb_960_h?: string, thumb_960_w?: string, thumb_gif?: string, thumb_pdf?: string, thumb_pdf_h?: string, thumb_pdf_w?: string, thumb_tiny?: string, thumb_video?: string, thumb_video_h?: number, thumb_video_w?: number, timestamp?: number, title?: string, title_blocks?: unknown[], to?: unknown[], transcription?: unknown, update_notification?: number, updated?: number, url_private?: string, url_private_download?: string, url_static_preview?: string, user?: string, user_team?: string, username?: string, vtt?: string }[], hidden?: boolean, icons?: { emoji?: string, image_36?: string, image_48?: string, image_64?: string, image_72?: string }, inviter?: string, is_locked?: boolean, language?: { is_reliable?: boolean, locale?: string }, latest_reply?: string, metadata?: { event_payload?: unknown, event_type?: string }, parent_user_id?: string, purpose?: string, reactions?: { count?: number, name?: string, url?: string, users?: string[] }[], reply_count?: number, reply_users?: string[], reply_users_count?: number, root?: { bot_id?: string, icons?: unknown, latest_reply?: string, parent_user_id?: string, reply_count?: number, reply_users?: string[], reply_users_count?: number, subscribed?: boolean, subtype?: string, text?: string, thread_ts?: string, ts?: string, type?: string, username?: string }, subscribed?: boolean, subtype?: string, team?: string, text?: string, thread_ts?: string, topic?: string, ts?: string, type?: string, upload?: boolean, user?: string, username?: string, x_files?: string[] } }[] }`\n\n#### generate3dModel\nGenerate a 3D model using a 3D generation model.\n- Text-to-3D models use the prompt field.\n- Image-to-3D and multi-view models take image URLs through the selected model\'s configuration inputs.\n- The output is standardized as a GLB URL plus optional FBX/OBJ/USDZ, thumbnail, texture maps, and provider task ID.\n- Input: `{ prompt?: string, intermediateAsset?: boolean, threeDModelOverride?: { model: string, config?: object } }`\n- Output: `{ prompt?: string, glbUrl: string, fbxUrl?: string, objUrl?: string, usdzUrl?: string, thumbnailUrl?: string, textureUrls?: object[], providerTaskId?: string, resolvedConfig?: object }`\n\n#### generateAsset\nGenerate an HTML asset and export it as a webpage, PDF, or image\n- Agents can generate HTML documents and export as webpage, PDFs, images, or videos. They do this by using the "generatePdf" block, which defines an HTML page with variables, and then the generation process renders the page to create the output and save its URL at the specified variable.\n- The template for the HTML page is generated by a separate process, and it can only use variables that have already been defined in the workflow at the time of its execution. It has full access to handlebars to render the HTML template, including a handlebars helper to render a markdown variable string as HTML (which can be useful for creating templates that render long strings). The template can also create its own simple JavaScript to do things like format dates and strings.\n- If PDF or composited image generation are part of the workflow, assistant adds the block and leaves the "source" empty. In a separate step, assistant generates a detailed request for the developer who will write the HTML.\n- Can also auto-generate HTML from a prompt (like a generate text block to generate HTML). In these cases, create a prompt with variables in the dynamicPrompt variable describing, in detail, the document to generate\n- Can either display output directly to user (foreground mode) or save the URL of the asset to a variable (background mode)\n- Input: `{ source: string, sourceType: "html" | "markdown" | "spa" | "raw" | "dynamic" | "customInterface", outputFormat: "pdf" | "png" | "html" | "mp4" | "openGraph", pageSize: "full" | "letter" | "A4" | "custom", testData: object, options?: { pageWidthPx?: number, pageHeightPx?: number, pageOrientation?: "portrait" | "landscape", rehostMedia?: boolean, videoDurationSeconds?: number }, spaSource?: { source?: string, lastCompiledSource?: string, files?: object, paths: string[], root: string, zipUrl: string }, rawSource?: string, dynamicPrompt?: string, dynamicSourceModelOverride?: { model: string, temperature: number, maxResponseTokens: number, ignorePreamble?: boolean, userMessagePreprocessor?: { dataSource?: string, messageTemplate?: string, maxResults?: number, enabled?: boolean, shouldInherit?: boolean }, preamble?: string, multiModelEnabled?: boolean, editResponseEnabled?: boolean, promptCaching?: boolean, config?: object }, transitionControl?: "default" | "native", shareControl?: "default" | "hidden", shareImageUrl?: string, intermediateAsset?: boolean }`\n- Output: `{ url: string }`\n\n#### generateImage\nGenerate an image from a text prompt using an AI model.\n- Prompts should be descriptive but concise (roughly 3\u20136 sentences).\n- Images are hosted on the public CDN by default, or in one of the app\'s own file stores when the call passes `store` \u2014 a private store keeps the image private.\n- In foreground mode, the image is displayed to the user. In background mode, the URL is saved to a variable.\n- When generateVariants is true with numVariants > 1, multiple images are generated in parallel.\n- In direct execution, foreground mode behaves as background, and userSelect variant behavior behaves as saveAll.\n- Input: `{ prompt: string, intermediateAsset?: boolean, imageModelOverride?: { model: string, config?: object }, generateVariants?: boolean, numVariants?: number, addWatermark?: boolean }`\n- Output: `{ imageUrl: string | string[] }`\n\n#### generateLipsync\nGenerate a lip sync video from provided audio and image.\n- In foreground mode, the video is displayed to the user. In background mode, the URL is saved to a variable.\n- Input: `{ intermediateAsset?: boolean, addWatermark?: boolean, lipsyncModelOverride?: { model: string, config?: object } }`\n- Output: `unknown`\n\n#### generateMusic\nGenerate an audio file from provided instructions (text) using a music model.\n- The text field contains the instructions (prompt) for the music generation.\n- In foreground mode, the audio is displayed to the user. In background mode, the URL is saved to a variable.\n- Input: `{ text: string, intermediateAsset?: boolean, musicModelOverride?: { model: string, config?: object } }`\n- Output: `unknown`\n\n#### generateSound\nGenerate a sound effect from a text description using a sound model.\n- The text field contains the description of the sound effect to generate.\n- In foreground mode, the audio is displayed to the user. In background mode, the URL is saved to a variable.\n- Input: `{ text: string, intermediateAsset?: boolean, soundModelOverride?: { model: string, config?: object } }`\n- Output: `{ audioUrl: string }`\n\n#### generateStaticVideoFromImage\nConvert a static image to an MP4\n- Can use to create slides/intertitles/slates for video composition\n- Input: `{ imageUrl: string, duration: string }`\n- Output: `{ videoUrl: string }`\n\n#### generateText\nSend a message to an AI model and return the response, or echo a system message.\n- Source "user" sends the message to an LLM and returns the model\'s response.\n- Source "system" echoes the message content directly (no AI call).\n- Mode "background" saves the result to a variable. Mode "foreground" streams it to the user (not available in direct execution).\n- Structured output (JSON/CSV) can be enforced via structuredOutputType and structuredOutputExample.\n- When executed inside a v2 app method (managed sandbox or local dev tunnel),\nLLM token output can be streamed to the frontend in real time via an SSE\nside-channel. The frontend opts in by passing { stream: true } to the method\ninvocation via @mindstudio-ai/interface. Tokens are published to Redis\npub/sub as they arrive and forwarded as SSE events on the invoke response.\nThe method code itself is unchanged \u2014 streaming is transparent to the\ndeveloper. See Sandboxes/execute.ts and the invoke handler in V2Apps for\nthe server-side plumbing.\n- Input: `{ message: string, source?: "user" | "system", modelOverride?: { model: string, temperature: number, maxResponseTokens: number, ignorePreamble?: boolean, userMessagePreprocessor?: { dataSource?: string, messageTemplate?: string, maxResults?: number, enabled?: boolean, shouldInherit?: boolean }, preamble?: string, multiModelEnabled?: boolean, editResponseEnabled?: boolean, promptCaching?: boolean, config?: object }, structuredOutputType?: "text" | "json" | "csv", structuredOutputExample?: string, chatHistoryMode?: "include" | "exclude" }`\n- Output: `{ content: string }`\n\n#### generateVideo\nGenerate a video from a text prompt using an AI model.\n- Prompts should be descriptive but concise (roughly 3\u20136 sentences).\n- Videos are hosted on the public CDN by default, or in one of the app\'s own file stores when the call passes `store` \u2014 a private store keeps the video private, with no public copy left behind.\n- In foreground mode, the video is displayed to the user. In background mode, the URL is saved to a variable.\n- When generateVariants is true with numVariants > 1, multiple videos are generated in parallel.\n- In direct execution, foreground mode behaves as background, and userSelect variant behavior behaves as saveAll.\n- Input: `{ prompt: string, intermediateAsset?: boolean, videoModelOverride?: { model: string, config?: object }, generateVariants?: boolean, numVariants?: number, addWatermark?: boolean }`\n- Output: `{ videoUrl: string | string[] }`\n\n#### getGmailAttachments\nDownload attachments from a Gmail email and store them on the app\'s own storage.\n- Requires a Google OAuth connection with Gmail readonly scope.\n- Attachments are stored and returned as URLs \u2014 on the public CDN by default, or in one of the app\'s own file stores when the call passes `store`. Use a private store for anything confidential; a public CDN URL is readable by anyone who has it.\n- Attachments larger than 25MB are skipped.\n- Use the message ID from Search Gmail Emails, List Recent Gmail Emails, or Get Gmail Email steps.\n- Input: `{ messageId: string, connectionId?: string }`\n- Output: `unknown`\n\n#### getGmailUnreadCount\nGet the number of unread emails in the connected Gmail inbox.\n- Requires a Google OAuth connection with Gmail readonly scope.\n- Returns the unread message count for the inbox label.\n- This is a lightweight call that does not fetch any email content.\n- Input: `{ connectionId?: string }`\n- Output: `unknown`\n\n#### getMediaMetadata\nGet info about a media file\n- Input: `{ mediaUrl: string }`\n- Output: `{ metadata: string }`\n\n#### imageFaceSwap\nReplace a face in an image with a face from another image using AI.\n- Requires both a target image and a face source image.\n- Output is a PNG on the public CDN, or in one of the app\'s own file stores when the call passes `store` \u2014 a private store keeps it private.\n- Input: `{ imageUrl: string, faceImageUrl: string, engine: string }`\n- Output: `{ imageUrl: string }`\n\n#### imageRemoveWatermark\nRemove watermarks from an image using AI.\n- Output is a PNG on the public CDN, or in one of the app\'s own file stores when the call passes `store` \u2014 a private store keeps it private.\n- Input: `{ imageUrl: string, engine: string, intermediateAsset?: boolean }`\n- Output: `{ imageUrl: string }`\n\n#### insertVideoClips\nInsert b-roll clips into a base video at a timecode, optionally with an xfade transition.\n- Input: `{ baseVideoUrl: string, overlayVideos: { videoUrl: string, startTimeSec: number }[], transition?: string, transitionDuration?: number, useOverlayAudio?: boolean, intermediateAsset?: boolean }`\n- Output: `{ videoUrl: string }`\n\n#### listDataSources\nLegacy v1 data sources. v2 apps should use `dataSources.defineDataSource()` from @mindstudio-ai/agent instead, which owns parsing, chunking, embedding and citations. List all data sources for the current app.\n- Returns metadata for every data source associated with the current app version.\n- Each entry includes the data source ID, name, description, status, and document list.\n- Input: `object`\n- Output: `unknown`\n\n#### listGmailLabels\nList all labels in the connected Gmail account. Use these label IDs or names with the Update Gmail Labels step.\n- Requires a Google OAuth connection with Gmail readonly scope.\n- Returns both system labels (INBOX, SENT, TRASH, etc.) and user-created labels.\n- Label type is "system" for built-in labels or "user" for custom labels.\n- Input: `{ connectionId?: string }`\n- Output: `unknown`\n\n#### listRecentGmailEmails\nList recent emails from the connected Gmail inbox.\n- Requires a Google OAuth connection with Gmail readonly scope.\n- Returns up to 100 emails (default 5), ordered by most recent first.\n- Functionally equivalent to Search Gmail Emails with an "in:inbox" query.\n- Input: `{ connectionId?: string, exportType: "json" | "text", limit: string }`\n- Output: `unknown`\n\n#### logic\nRoute execution to different branches based on AI evaluation, comparison operators, or workflow jumps.\n- Supports two modes: "ai" (default) uses an AI model to pick the most accurate statement; "comparison" uses operator-based checks.\n- In AI mode, the model picks the most accurate statement from the list. All possible cases must be specified.\n- In comparison mode, the context is the left operand and each case\'s condition is the right operand. First matching case wins. Use operator "default" as a fallback.\n- Requires at least two cases.\n- Each case can transition to a step in the current workflow (destinationStepId) or jump to another workflow (destinationWorkflowId).\n- Input: `{ mode?: "ai" | "comparison", context: string, cases: ({ id: string, condition: string, operator?: "eq" | "neq" | "gt" | "lt" | "gte" | "lte" | "exists" | "not_exists" | "contains" | "not_contains" | "default", destinationStepId?: string, destinationWorkflowId?: string } | string)[], modelOverride?: { model: string, temperature: number, maxResponseTokens: number, ignorePreamble?: boolean, userMessagePreprocessor?: { dataSource?: string, messageTemplate?: string, maxResults?: number, enabled?: boolean, shouldInherit?: boolean }, preamble?: string, multiModelEnabled?: boolean, editResponseEnabled?: boolean, promptCaching?: boolean, config?: object } }`\n- Output: `{ selectedCase: number }`\n\n#### makeDotComRunScenario\nTrigger a Make.com (formerly Integromat) scenario via webhook and return the response.\n- The webhook URL must be configured in your Make.com scenario.\n- Input key-value pairs are sent as JSON in the POST body.\n- Response format depends on the Make.com scenario configuration.\n- Input: `{ webhookUrl: string, input: object }`\n- Output: `{ data: unknown }`\n\n#### mergeAudio\nMerge one or more clips into a single audio file.\n- Input: `{ mp3Urls: string[], fileMetadata?: object, albumArtUrl?: string, intermediateAsset?: boolean }`\n- Output: `{ audioUrl: string }`\n\n#### mergeVideos\nMerge one or more clips into a single video.\n- Input: `{ videoUrls: string[], transition?: string, transitionDuration?: number, intermediateAsset?: boolean }`\n- Output: `{ videoUrl: string }`\n\n#### mixAudioIntoVideo\nMix an audio track into a video\n- Input: `{ videoUrl: string, audioUrl: string, options: { keepVideoAudio?: boolean, audioGainDb?: number, videoGainDb?: number, loopAudio?: boolean }, intermediateAsset?: boolean }`\n- Output: `{ videoUrl: string }`\n\n#### muteVideo\nMute a video file\n- Input: `{ videoUrl: string, intermediateAsset?: boolean }`\n- Output: `{ videoUrl: string }`\n\n#### n8nRunNode\nTrigger an n8n workflow node via webhook and return the response.\n- The webhook URL must be configured in your n8n workflow.\n- Supports GET and POST methods with optional Basic authentication.\n- For GET requests, input values are sent as query parameters. For POST, they are sent as JSON body.\n- Input: `{ method: string, authentication: "none" | "basic" | "string", user: string, password: string, webhookUrl: string, input: object }`\n- Output: `{ data: unknown }`\n\n#### postToSlackChannel\nSend a message to a Slack channel via a connected bot.\n- The user is responsible for connecting their Slack workspace and selecting the channel\n- Supports both simple text messages and slack blocks messages\n- Text messages can use limited markdown (slack-only fomatting\u2014e.g., headers are just rendered as bold)\n- Input: `{ channelId: string, messageType: "string" | "blocks", message: string, connectionId?: string }`\n- Output: `unknown`\n\n#### postToZapier\nSend data to a Zapier Zap via webhook and return the response.\n- The webhook URL must be configured in the Zapier Zap settings\n- Input keys and values are sent as the JSON body of the POST request\n- The webhook response (JSON or plain text) is returned as the output\n- Input: `{ webhookUrl: string, input: object }`\n- Output: `{ data: unknown }`\n\n#### queryDataSource\nLegacy v1 data sources. v2 apps should use `dataSources.defineDataSource()` from @mindstudio-ai/agent instead, which owns parsing, chunking, embedding and citations. Search a vector data source (RAG) and return relevant document chunks.\n- Queries a vectorized data source and returns the most relevant chunks.\n- Useful for retrieval-augmented generation (RAG) workflows.\n- Input: `{ dataSourceId: string, query: string, maxResults: number }`\n- Output: `{ text: string, chunks: string[], query: string, citations: unknown[], latencyMs: number }`\n\n#### queryExternalDatabase\nExecute a SQL query against an external database connected to the workspace.\n- Requires a database connection configured in the workspace.\n- Supports PostgreSQL (including Supabase), MySQL, and MSSQL.\n- Results can be returned as JSON or CSV.\n- Input: `{ connectionId?: string, query: string, outputFormat: "json" | "csv" }`\n- Output: `{ data: unknown }`\n\n#### redactPII\nReplace personally identifiable information in text with placeholders using Microsoft Presidio.\n- PII is replaced with entity type placeholders (e.g. "Call me at <PHONE_NUMBER>").\n- If entities is empty, returns empty text immediately without processing.\n- Input: `{ input: string, language: string, entities: string[] }`\n- Output: `{ text: string }`\n\n#### removeBackgroundFromImage\nRemove the background from an image using AI, producing a transparent PNG.\n- Uses the Bria background removal model via fal.ai by default.\n- Uses WaveSpeed\'s Ideogram background removal model when type is "advanced".\n- Output is a PNG with transparency on the public CDN, or in one of the app\'s own file stores when the call passes `store` \u2014 a private store keeps it private, `autoCrop` included.\n- Input: `{ type?: "standard" | "advanced", imageUrl: string, autoCrop?: boolean }`\n- Output: `{ imageUrl: string }`\n\n#### resizeVideo\nResize a video file\n- Input: `{ videoUrl: string, mode: "fit" | "exact", maxWidth?: number, maxHeight?: number, width?: number, height?: number, strategy?: "pad" | "crop", intermediateAsset?: boolean }`\n- Output: `{ videoUrl: string }`\n\n#### runFromConnectorRegistry\nRun a raw API connector to a third-party service\n- Use the /developer/v2/helpers/connectors endpoint to list available services and actions.\n- Use /developer/v2/helpers/connectors/{serviceId}/{actionId} to get the full input configuration for an action.\n- Use /developer/v2/helpers/connections to list your available OAuth connections.\n- The actionId format is "serviceId/actionId" (e.g., "slack/send-message").\n- Pass a __connectionId to authenticate the request with a specific OAuth connection, otherwise the default will be used (if configured).\n- Input: `{ actionId: string, displayName: string, icon: string, configurationValues: object, __connectionId?: string }`\n- Output: `{ data: object }`\n\n#### scrapeLinkedInCompany\nScrape public company data from a LinkedIn company page.\n- Requires a LinkedIn company URL (e.g. https://www.linkedin.com/company/mindstudioai).\n- Returns structured company data including description, employees, updates, and similar companies.\n- Input: `{ url: string }`\n- Output: `{ company: unknown }`\n\n#### scrapeLinkedInProfile\nScrape public profile data from a LinkedIn profile page.\n- Requires a LinkedIn profile URL (e.g. https://www.linkedin.com/in/username).\n- Returns structured profile data including experience, education, articles, and activities.\n- Input: `{ url: string }`\n- Output: `{ profile: unknown }`\n\n#### scrapeUrl\nExtract text, HTML, or structured content from one or more web pages.\n- Accepts a single URL or multiple URLs (as a JSON array, comma-separated, or newline-separated).\n- Output format controls the result shape: "text" returns markdown, "html" returns raw HTML, "json" returns structured scraper data, "summary" returns a model-written summary and requires the "firecrawl" service.\n- Can optionally capture a screenshot of each page. Screenshots go to the public CDN by default, or into one of the app\'s own file stores when the call passes `store`.\n- Handles bot protection automatically; no proxy or rendering configuration is needed.\n- Input: `{ url: string, service?: "default" | "firecrawl", autoEnhance?: boolean, outputFormat?: "text" | "json" | "html" | "summary", pageOptions?: { onlyMainContent: boolean, screenshot: boolean, waitFor: number, replaceAllPathsWithAbsolutePaths: boolean, headers: object, removeTags: string[], mobile: boolean } }`\n- Output: `{ content: string | string[] | { text: string, html: string, json?: object, screenshotUrl?: string, metadata?: { title: string, description: string, url: string, image: string }, error?: { code: string, message: string }, costUnits?: number } | { text: string, html: string, json?: object, screenshotUrl?: string, metadata?: { title: string, description: string, url: string, image: string }, error?: { code: string, message: string }, costUnits?: number }[], screenshot?: string | string[] }`\n\n#### scrapeXPost\nScrape data from a single X (Twitter) post by URL.\n- Returns structured post data (text, html, optional json/screenshot/metadata).\n- Optionally saves the text content to a variable.\n- Input: `{ url: string }`\n- Output: `{ post: { text: string, html: string, json?: object, screenshotUrl?: string, metadata?: { title: string, description: string, url: string, image: string }, error?: { code: string, message: string }, costUnits?: number } }`\n\n#### scrapeXProfile\nScrape public profile data from an X (Twitter) account by URL.\n- Returns structured profile data.\n- Optionally saves the result to a variable.\n- Input: `{ url: string }`\n- Output: `{ profile: { text: string, html: string, json?: object, screenshotUrl?: string, metadata?: { title: string, description: string, url: string, image: string }, error?: { code: string, message: string }, costUnits?: number } }`\n\n#### screenshotUrl\nCapture a screenshot of a web page as a PNG image.\n- Takes a viewport or full-page screenshot of the given URL.\n- Returns a PNG image URL: on the public CDN by default, or in one of the app\'s own file stores when the call passes `store` \u2014 a private store keeps the screenshot private.\n- Viewport mode captures only the visible area; fullPage captures the entire scrollable page, scrolling it first so lazy-loaded content is included.\n- Works on sites behind bot protection (Cloudflare and similar).\n- Cookie-consent banners and similar overlays are suppressed automatically.\n- `width` and `height` are the exact output pixel dimensions.\n- You can add a delay or wait for a CSS selector before capturing. A selector that never appears fails the step rather than capturing early.\n- Input: `{ url: string, mode?: "viewport" | "fullPage", width?: number, height?: number, delay?: number, waitFor?: string }`\n- Output: `{ screenshotUrl: string }`\n\n#### searchGmailEmails\nSearch for emails in the connected Gmail account using a Gmail search query. To list recent inbox emails, pass an empty query string.\n- Requires a Google OAuth connection with Gmail readonly scope.\n- Uses Gmail search syntax (e.g. "from:user@example.com", "subject:invoice", "is:unread").\n- To list recent inbox emails, use an empty query string or "in:inbox".\n- Returns up to 100 emails (default 5). The variable receives text or JSON depending on exportType.\n- The direct execution output always returns structured email objects.\n- Input: `{ query: string, connectionId?: string, exportType: "json" | "text", limit: string }`\n- Output: `{ emails: { id: string, subject: string, from: string, to: string, date: string, plainBody: string, htmlBody: string, labels: string }[] }`\n\n#### searchGoogle\nSearch the web using Google and return structured results.\n- Defaults to us/english, but can optionally specify country and/or language.\n- Defaults to any time, but can optionally specify last hour, last day, week, month, or year.\n- Defaults to top 30 results, but can specify 1 to 100 results to return.\n- Set fetchTopN to also read the page content of the first N results in the same step, instead of following up with a separate Scrape URL step per result. Costs meaningfully more (each page is a full scrape), so ask for the fewest you need.\n- Also returns Google\'s "people also ask" and related searches when available, which are useful for choosing a follow-up query.\n- Input: `{ query: string, exportType: "text" | "json", countryCode?: string, languageCode?: string, dateRange?: "hour" | "day" | "week" | "month" | "year" | "any", numResults?: number, fetchTopN?: number }`\n- Output: `{ results: { title: string, description: string, url: string, content?: string }[], peopleAlsoAsk?: { question?: string, snippet?: string, link?: string }[], relatedSearches?: string[] }`\n\n#### searchGoogleImages\nSearch Google Images and return image results with URLs and metadata.\n- Defaults to us/english, but can optionally specify country and/or language.\n- Defaults to any time, but can optionally specify last hour, last day, week, month, or year.\n- Defaults to top 30 results, but can specify 1 to 100 results to return.\n- Input: `{ query: string, exportType: "text" | "json", countryCode?: string, languageCode?: string, dateRange?: "hour" | "day" | "week" | "month" | "year" | "any", numResults?: number }`\n- Output: `{ images: { title: string, imageUrl: string, imageWidth: number, imageHeight: number, thumbnailUrl: string, thumbnailWidth: number, thumbnailHeight: number, source: string, domain: string, link: string, googleUrl: string, position: number }[] }`\n\n#### searchGoogleNews\nSearch Google News for recent news articles matching a query.\n- Defaults to top 30 results, but can specify 1 to 100 results to return.\n- Input: `{ text: string, exportType: "text" | "json", numResults?: number }`\n- Output: `{ articles: { title: string, link: string, date: string, source: { name: string }, snippet?: string }[] }`\n\n#### searchGoogleTrends\nFetch Google Trends data for a search term.\n- date accepts shorthand ("now 1-H", "today 1-m", "today 5-y", etc.) or custom "yyyy-mm-dd yyyy-mm-dd" ranges.\n- data_type controls the shape of returned data: TIMESERIES, GEO_MAP, GEO_MAP_0, RELATED_TOPICS, or RELATED_QUERIES.\n- Input: `{ text: string, hl: string, geo: string, data_type: "TIMESERIES" | "GEO_MAP" | "GEO_MAP_0" | "RELATED_TOPICS" | "RELATED_QUERIES", cat: string, date: string, ts: string }`\n- Output: `{ trends: object }`\n\n#### searchPerplexity\nSearch the web using the Perplexity API and return structured results.\n- Defaults to US results. Use countryCode (ISO code) to filter by country.\n- Returns 10 results by default, configurable from 1 to 20.\n- The variable receives text or JSON depending on exportType. The direct execution output always returns structured results.\n- Input: `{ query: string, exportType: "text" | "json", countryCode?: string, numResults?: number }`\n- Output: `{ results: { title: string, description: string, url: string }[] }`\n\n#### sendEmail\nSend an email to one or more recipient addresses.\n- If you are writing a backend method for a v2 app, prefer mindstudio.email.send({...}) \u2014 it takes the same fields, returns the batchId you read stats back with, and sits alongside email.messages / email.batch / email.suppress. This action remains fully supported and runs the same code; it is also the only way to use connectionId or generateHtml.\n- Use the "to" field to send to one or more specific recipient email addresses directly. Allowed recipients depend on the sender: when the app sends from a domain it owns (a verified custom domain or its <slug>.madewithremy.com subdomain) any recipient is allowed; when it falls back to the shared Remy address, recipients must be verified app users or members of the app\'s organization. (v1 apps cannot use a direct "to" \u2014 they must resolve recipients via a connection.)\n- Alternatively, recipient email addresses can be resolved from OAuth connections configured by the app creator via connectionId. The user running the workflow does not specify the recipient directly.\n- Use "cc" and "bcc" to add visible / hidden recipients (a string or an array). They are subject to the same recipient rules as "to".\n- "to" is optional only in the sense that recipients can come from elsewhere \u2014 omit it and supply "cc"/"bcc" for a hidden-list send, or omit all three and recipients are resolved from an OAuth connection. Naming nobody at all is an error.\n- Bcc-only sends are supported: with no "to" or "cc", the To: header is addressed to the app\'s own sender address (the standard "undisclosed recipients" pattern) so recipients can\'t see each other. The returned "recipients" reflects that auto-filled address.\n- Every recipient counts toward the app\'s daily outbound cap, including cc and bcc.\n- The sender defaults automatically: v2 apps send from the app\'s own identity \u2014 its verified custom domain if set, else its platform subdomain (noreply@<slug>.madewithremy.com), else the default Remy address.\n- Optionally set "from" to a custom handle, but ONLY if the app has a custom domain or subdomain: a bare handle ("support" \u2192 support@<app-domain>), a full "support@your-domain.com", or "Name <support@your-domain.com>". The domain must be one the app owns, or the step fails.\n- If the body is a URL to a hosted HTML file on the CDN, the HTML is fetched and used as the email body.\n- The body is interpreted automatically: if it already looks like HTML it is sent as HTML, otherwise it is rendered from Markdown. Every email is sent as multipart with a plain-text alternative auto-derived from the body (better deliverability).\n- Set bodyType to override interpretation: "html" (send as-is), "markdown" (render to HTML), or "text" (plain text only, no HTML part). Default is "auto".\n- Optionally set "text" to supply your own plain-text alternative instead of the auto-derived one.\n- Use "attachments" to attach files by URL. Each entry is a URL string, or an object { url, filename?, contentType? } to control the attachment\'s displayed filename and MIME type.\n- When generateHtml is enabled, the body text is converted to a styled HTML email using an AI model (implies an HTML body).\n- Set replyTo to control the Reply-To address for replies.\n- For threaded replies in a shared inbox, set inReplyTo (the Message-ID being replied to) and references (prior Message-IDs in the thread).\n- connectionId can be a comma-separated list to send to multiple recipients.\n- The special connectionId "trigger_email" uses the email address that triggered the workflow.\n- Sends default to category "marketing": every message carries a one-click unsubscribe header pointing at the app\'s own domain, and recipients who previously unsubscribed from this app are skipped and returned in "suppressed". Because the unsubscribe link is per-recipient, a marketing send delivers one separate message per recipient and cc/bcc are folded into that list.\n- Set category to "transactional" for receipts, alerts, password resets and anything a recipient must not be able to opt out of. Transactional sends carry no unsubscribe header, ignore the unsubscribe list, and keep cc/bcc on a single message \u2014 use it whenever cc/bcc semantics matter.\n- The subject is stored verbatim in the app\'s delivery log, which every workspace member can read for 90 days. Never put a secret in the subject \u2014 no one-time codes, no magic-link tokens, no password-reset tokens. Put those in the body of a "transactional" send: a small transactional send is delivered immediately and its body is never stored. A "marketing" send, or any large send, is queued for delivery, which means its body is held in internal storage until the send finishes \u2014 so do not put a secret in the body of one of those either.\n- Input: `{ subject: string, body: string, to?: string | string[], cc?: string | string[], bcc?: string | string[], connectionId?: string, generateHtml?: boolean, generateHtmlInstructions?: string, generateHtmlModelOverride?: { model: string, temperature: number, maxResponseTokens: number, ignorePreamble?: boolean, userMessagePreprocessor?: { dataSource?: string, messageTemplate?: string, maxResults?: number, enabled?: boolean, shouldInherit?: boolean }, preamble?: string, multiModelEnabled?: boolean, editResponseEnabled?: boolean, promptCaching?: boolean, config?: object }, attachments?: (string | { url: string, filename?: string, contentType?: string })[], from?: string, bodyType?: "auto" | "html" | "markdown" | "text", text?: string, replyTo?: string, inReplyTo?: string, references?: string[], category?: "transactional" | "marketing", batchId?: string }`\n- Output: `{ recipients: string[], cc: string[], bcc: string[], from: string, suppressed: string[], batchId: string }`\n\n#### sendGmailDraft\nSend an existing draft from the connected Gmail account.\n- Requires a Google OAuth connection with Gmail compose scope.\n- The draft is sent and removed from the Drafts folder.\n- Use the draft ID returned by the Create Gmail Draft or List Gmail Drafts steps.\n- Input: `{ draftId: string, connectionId?: string }`\n- Output: `unknown`\n\n#### sendGmailMessage\nSend an email from the connected Gmail account.\n- Requires a Google OAuth connection with Gmail compose scope.\n- messageType controls the body format: "plain" for plain text, "html" for raw HTML, "markdown" for auto-converted markdown.\n- Input: `{ to: string, subject: string, message: string, connectionId?: string, messageType: "plain" | "html" | "markdown" }`\n- Output: `{ messageId: string }`\n\n#### sendSlackDirectMessage\nSend a direct message to a Slack user via a connected bot.\n- The user is responsible for connecting their Slack workspace\n- The recipient is identified by their Slack user ID\n- Supports both simple text messages and Slack blocks messages\n- Text messages can use limited markdown (slack-only formatting\u2014e.g., headers are just rendered as bold)\n- Input: `{ slackUserId: string, messageType: "string" | "blocks", message: string, connectionId?: string }`\n- Output: `unknown`\n\n#### sendSMS\nSend an SMS or MMS message to a phone number configured via OAuth connection.\n- User is responsible for configuring the connection to the number (MindStudio requires double opt-in to prevent spam)\n- If mediaUrls are provided, the message is sent as MMS instead of SMS\n- MMS supports up to 10 media URLs (images, video, audio, PDF) with a 5MB limit per file\n- MMS is only supported on US and Canadian carriers; international numbers will receive SMS only (media silently dropped)\n- Input: `{ body: string, connectionId?: string, mediaUrls?: string[] }`\n- Output: `unknown`\n\n#### setGmailReadStatus\nMark one or more Gmail emails as read or unread.\n- Requires a Google OAuth connection with Gmail modify scope.\n- Accepts one or more message IDs as a comma-separated string or array.\n- Set markAsRead to true to mark as read, false to mark as unread.\n- Input: `{ messageIds: string, markAsRead: boolean, connectionId?: string }`\n- Output: `unknown`\n\n#### setRunTitle\nSet the title of the agent run for the user\'s history\n- Input: `{ title: string }`\n- Output: `unknown`\n\n#### setVariable\nExplicitly set a variable to a given value.\n- Useful for bootstrapping global variables or setting constants.\n- The variable name and value both support variable interpolation.\n- The type field is a UI hint only (controls input widget in the editor).\n- Input: `{ value: string | string[] }`\n- Output: `object`\n\n#### textToSpeech\nGenerate an audio file from provided text using a speech model.\n- The text field contains the exact words to be spoken (not instructions).\n- In foreground mode, the audio is displayed to the user. In background mode, the URL is saved to a variable.\n- Input: `{ text: string, intermediateAsset?: boolean, speechModelOverride?: { model: string, config?: object } }`\n- Output: `{ audioUrl: string }`\n\n#### transcribeAudio\nConvert an audio file to text using a transcription model.\n- The prompt field provides optional context to improve transcription accuracy (e.g. language, speaker names, domain).\n- Detail beyond the plain transcript is requested through `transcriptionModelOverride.config`, and support varies by model:\n- `includeSpeakers: \'yes\'` \u2014 label each passage with a speaker. Supported by elevenlabs-scribe-v1, elevenlabs-scribe-v2, deepgram-nova-3 and grok-stt. Also switches `text` to a timestamped, speaker-labelled block format.\n- `includeWordTimings: \'yes\'` \u2014 add per-word timings and confidence inside each segment. Supported by elevenlabs-scribe-v1, elevenlabs-scribe-v2, deepgram-nova-3 and whisper-1. Off by default because an hour of speech is roughly ten thousand words.\n- `detectLanguage: \'yes\'` \u2014 detect the spoken language instead of assuming English (deepgram-nova-3 only; Scribe and the whisper models always report it).\n- The `segments` output carries timings, confidence and speakers as structured data, so there is no need to slice the audio up to recover them. It is populated for elevenlabs-scribe-v1/v2, deepgram-nova-3, whisper-1 and the DeepInfra-served whisper/voxtral/qwen models, and absent for models that return nothing but text (gpt-transcribe, gpt-4o-transcribe, grok-stt).\n- `confidence` is always 0..1 with higher meaning more confident, whichever model produced it.\n- Input: `{ audioUrl: string, prompt: string, transcriptionModelOverride?: { model: string, config?: object } }`\n- Output: `{ text: string, segments?: { start: number, end: number, text: string, confidence?: number, speaker?: string, language?: string, words?: { text: string, start?: number, end?: number, confidence?: number, speaker?: string }[] }[], language?: string, durationSeconds?: number }`\n\n#### trimMedia\nTrim an audio or video clip\n- Input: `{ inputUrl: string, start?: number | string, duration?: string | number, intermediateAsset?: boolean }`\n- Output: `{ mediaUrl: string }`\n\n#### updateGmailLabels\nAdd or remove labels on Gmail messages, identified by message IDs or a search query.\n- Requires a Google OAuth connection with Gmail modify scope.\n- Provide either a query (Gmail search syntax) or explicit messageIds to target messages.\n- Label IDs can be label names or Gmail label IDs \u2014 names are resolved automatically.\n- Input: `{ query: string, connectionId?: string, messageIds: string, addLabelIds: string, removeLabelIds: string }`\n- Output: `{ updatedMessageIds: string[] }`\n\n#### uploadDataSourceDocument\nLegacy v1 data sources. v2 apps should use `dataSources.defineDataSource()` from @mindstudio-ai/agent instead, which owns parsing, chunking, embedding and citations. Upload a file into an existing data source from a URL or raw text content.\n- If "file" is a single URL, the file is downloaded from that URL and uploaded.\n- If "file" is any other string, a .txt document is created from that content and uploaded.\n- The block waits (polls) for processing to complete before transitioning, up to 5 minutes.\n- Once processing finishes, vectors are loaded into Milvus so the data source is immediately queryable.\n- Supported file types (when using a URL) are the same as the data source upload UI (PDF, DOCX, TXT, etc.).\n- Input: `{ dataSourceId: string, file: string, fileName: string }`\n- Output: `unknown`\n\n#### upscaleImage\nIncrease the resolution of an image using AI upscaling.\n- Output is a PNG on the public CDN, or in one of the app\'s own file stores when the call passes `store` \u2014 a private store keeps it private.\n- Input: `{ imageUrl: string, targetResolution: "2k" | "4k" | "8k", engine: "standard" | "pro" }`\n- Output: `{ imageUrl: string }`\n\n#### upscaleVideo\nUpscale a video file\n- Input: `{ videoUrl: string, targetResolution: "720p" | "1080p" | "2K" | "4K", engine: "standard" | "pro" | "ultimate" | "flashvsr" | "seedance" | "seedvr2" | "runwayml/upscale-v1", intermediateAsset?: boolean }`\n- Output: `{ videoUrl: string }`\n\n#### videoFaceSwap\nSwap faces in a video file\n- Input: `{ videoUrl: string, faceImageUrl: string, targetIndex: number, engine: string, intermediateAsset?: boolean }`\n- Output: `{ videoUrl: string }`\n\n#### videoRemoveBackground\nRemove or replace background from a video\n- Input: `{ videoUrl: string, newBackground: "transparent" | "image", newBackgroundImageUrl?: string, engine: string, intermediateAsset?: boolean }`\n- Output: `{ videoUrl: string }`\n\n#### videoRemoveWatermark\nRemove a watermark from a video\n- Input: `{ videoUrl: string, engine: string, intermediateAsset?: boolean }`\n- Output: `{ videoUrl: string }`\n\n#### watermarkImage\nOverlay a watermark image onto another image.\n- The watermark is placed at the specified corner with configurable padding and width.\n- Input: `{ imageUrl: string, watermarkImageUrl: string, corner: "top-left" | "top-right" | "bottom-left" | "bottom-right", paddingPx: number, widthPx: number, intermediateAsset?: boolean }`\n- Output: `{ imageUrl: string }`\n\n#### watermarkVideo\nAdd an image watermark to a video\n- Input: `{ videoUrl: string, imageUrl: string, corner: "top-left" | "top-right" | "bottom-left" | "bottom-right", paddingPx: number, widthPx: number, intermediateAsset?: boolean }`\n- Output: `{ videoUrl: string }`\n\n### ActiveCampaign\n\n#### activeCampaignAddNote\nAdd a note to an existing contact in ActiveCampaign.\n- Requires an ActiveCampaign OAuth connection (connectionId).\n- The contact must already exist \u2014 use the contact ID from a previous create or search step.\n- Input: `{ contactId: string, note: string, connectionId?: string }`\n- Output: `unknown`\n\n#### activeCampaignCreateContact\nCreate or sync a contact in ActiveCampaign.\n- Requires an ActiveCampaign OAuth connection (connectionId).\n- If a contact with the email already exists, it may be updated depending on ActiveCampaign settings.\n- Custom fields are passed as a key-value map where keys are field IDs.\n- Input: `{ email: string, firstName: string, lastName: string, phone: string, accountId: string, customFields: object, connectionId?: string }`\n- Output: `{ contactId: string }`\n\n### Airtable\n\n#### airtableCreateUpdateRecord\nCreate a new record or update an existing record in an Airtable table.\n- If recordId is provided, updates that record. Otherwise, creates a new one.\n- When updating with updateMode "onlySpecified", unspecified fields are left as-is. With "all", unspecified fields are cleared.\n- Array fields (e.g. multipleAttachments) accept arrays of values.\n- Input: `{ connectionId?: string, baseId: string, tableId: string, recordId?: string, updateMode?: "onlySpecified" | "all", fields: unknown, recordData: object }`\n- Output: `{ recordId: string }`\n\n#### airtableDeleteRecord\nDelete a record from an Airtable table by its record ID.\n- Requires an active Airtable OAuth connection (connectionId).\n- Silently succeeds if the record does not exist.\n- Input: `{ connectionId?: string, baseId: string, tableId: string, recordId: string }`\n- Output: `{ deleted: boolean }`\n\n#### airtableGetRecord\nFetch a single record from an Airtable table by its record ID.\n- Requires an active Airtable OAuth connection (connectionId).\n- If the record is not found, returns a string message instead of a record object.\n- Input: `{ connectionId?: string, baseId: string, tableId: string, recordId: string }`\n- Output: `{ record: { id: string, createdTime: string, fields: object } | null }`\n\n#### airtableGetTableRecords\nFetch multiple records from an Airtable table with optional pagination.\n- Requires an active Airtable OAuth connection (connectionId).\n- Default limit is 100 records. Maximum is 1000.\n- When outputFormat is \'csv\', the variable receives CSV text. The direct execution output always returns parsed records.\n- Input: `{ connectionId?: string, baseId: string, tableId: string, outputFormat?: "json" | "csv", limit?: number }`\n- Output: `{ records: { id: string, createdTime: string, fields: object }[] }`\n\n### Apollo\n\n#### enrichPerson\nLook up professional information about a person using Apollo.io. Search by ID, name, LinkedIn URL, email, or domain.\n- At least one search parameter must be provided.\n- Returns enriched data from Apollo including contact details, employment info, and social profiles.\n- Input: `{ params: { id: string, name: string, linkedinUrl: string, email: string, domain: string } }`\n- Output: `{ data: unknown }`\n\n#### peopleSearch\nSearch for people matching specific criteria using Apollo.io. Supports natural language queries and advanced filters.\n- Can use a natural language "smartQuery" which is converted to Apollo search parameters by an AI model.\n- Advanced params can override or supplement the smart query results.\n- Optionally enriches returned people and/or their organizations for additional detail.\n- Results are paginated. Use limit and page to control the result window.\n- Input: `{ smartQuery: string, enrichPeople: boolean, enrichOrganizations: boolean, limit: string, page: string, params: { personTitles: string, includeSimilarTitles: string, qKeywords: string, personLocations: string, personSeniorities: string, organizationLocations: string, qOrganizationDomainsList: string, contactEmailStatus: string, organizationNumEmployeesRanges: string, revenueRangeMin: string, revenueRangeMax: string, currentlyUsingAllOfTechnologyUids: string, currentlyUsingAnyOfTechnologyUids: string, currentlyNotUsingAnyOfTechnologyUids: string } }`\n- Output: `{ results: unknown }`\n\n### Coda\n\n#### codaCreateUpdatePage\nCreate a new page or update an existing page in a Coda document.\n- Requires a Coda OAuth connection (connectionId).\n- If pageData.pageId is provided, updates that page. Otherwise, creates a new one.\n- Page content is provided as markdown and converted to Coda\'s canvas format.\n- When updating, insertionMode controls how content is applied (default: \'append\').\n- Input: `{ connectionId?: string, pageData: { docId: string, pageId?: string, name: string, subtitle: string, iconName: string, imageUrl: string, parentPageId?: string, pageContent: string | unknown, contentUpdate?: unknown, insertionMode?: string } }`\n- Output: `{ pageId: string }`\n\n#### codaCreateUpdateRow\nCreate a new row or update an existing row in a Coda table.\n- Requires a Coda OAuth connection (connectionId).\n- If rowId is provided, updates that row. Otherwise, creates a new one.\n- Row data keys are column IDs. Empty values are excluded.\n- Input: `{ connectionId?: string, docId: string, tableId: string, rowId?: string, rowData: object }`\n- Output: `{ rowId: string }`\n\n#### codaFindRow\nSearch for a row in a Coda table by matching column values.\n- Requires a Coda OAuth connection (connectionId).\n- Returns the first row matching all specified column values, or null if no match.\n- Search criteria in rowData are ANDed together.\n- Input: `{ connectionId?: string, docId: string, tableId: string, rowData: object }`\n- Output: `{ row: { id: string, values: object } | null }`\n\n#### codaGetPage\nExport and read the contents of a page from a Coda document.\n- Requires a Coda OAuth connection (connectionId).\n- Page export is asynchronous on Coda\'s side \u2014 there may be a brief delay while it processes.\n- If a page was just created in a prior step, there is an automatic 20-second retry if the first export attempt fails.\n- Input: `{ connectionId?: string, docId: string, pageId: string, outputFormat?: "html" | "markdown" }`\n- Output: `{ content: string }`\n\n#### codaGetTableRows\nFetch rows from a Coda table with optional pagination.\n- Requires a Coda OAuth connection (connectionId).\n- Default limit is 10000 rows. Rows are fetched in pages of 500.\n- When outputFormat is \'csv\', the variable receives CSV text. The direct execution output always returns parsed rows.\n- Input: `{ connectionId?: string, docId: string, tableId: string, limit?: number | string, outputFormat?: "json" | "csv" }`\n- Output: `{ rows: { id: string, values: object }[] }`\n\n### Gmail\n\n#### deleteGmailEmail\nMove an email to trash in the connected Gmail account (recoverable delete).\n- Requires a Google OAuth connection with Gmail modify scope.\n- Uses trash (recoverable) rather than permanent delete.\n- Input: `{ messageId: string, connectionId?: string }`\n- Output: `unknown`\n\n#### getGmailDraft\nRetrieve a specific draft from Gmail by draft ID.\n- Requires a Google OAuth connection with Gmail readonly scope.\n- Returns the draft content including subject, recipients, sender, and body.\n- Input: `{ draftId: string, connectionId?: string }`\n- Output: `{ draftId: string, messageId: string, subject: string, to: string, from: string, body: string }`\n\n#### getGmailEmail\nRetrieve a specific email from Gmail by message ID.\n- Requires a Google OAuth connection with Gmail readonly scope.\n- Returns the email subject, sender, recipient, date, body (plain text preferred, falls back to HTML), and labels.\n- Input: `{ messageId: string, connectionId?: string }`\n- Output: `{ messageId: string, subject: string, from: string, to: string, date: string, body: string, labels: string }`\n\n#### listGmailDrafts\nList drafts in the connected Gmail account.\n- Requires a Google OAuth connection with Gmail readonly scope.\n- Returns up to 50 drafts (default 10).\n- The variable receives text or JSON depending on exportType.\n- Input: `{ connectionId?: string, limit?: string, exportType: "json" | "text" }`\n- Output: `{ drafts: { draftId: string, messageId: string, subject: string, to: string, snippet: string }[] }`\n\n#### replyToGmailEmail\nReply to an existing email in Gmail. The reply is threaded under the original message.\n- Requires a Google OAuth connection with Gmail compose and readonly scopes.\n- The reply is sent to the original sender and threaded under the original message.\n- messageType controls the body format: "plain", "html", or "markdown".\n- Input: `{ messageId: string, message: string, messageType: "plain" | "html" | "markdown", connectionId?: string }`\n- Output: `{ messageId: string }`\n\n### Google\n\n#### createGoogleDoc\nCreate a new Google Document and optionally populate it with content.\n- textType determines how the text field is interpreted: "plain" for plain text, "html" for HTML markup, "markdown" for Markdown.\n- Input: `{ title: string, text: string, connectionId?: string, textType: "plain" | "html" | "markdown" }`\n- Output: `{ documentUrl: string }`\n\n#### createGoogleSheet\nCreate a new Google Spreadsheet and populate it with CSV data.\n- Input: `{ title: string, text: string, connectionId?: string }`\n- Output: `{ spreadsheetUrl: string }`\n\n#### deleteGoogleSheetRows\nDelete a range of rows from a Google Spreadsheet.\n- Requires a Google OAuth connection with Drive scope.\n- startRow and endRow are 1-based row numbers (inclusive).\n- If sheetName is omitted, operates on the first sheet.\n- Input: `{ documentId: string, sheetName?: string, startRow: string, endRow: string, connectionId?: string }`\n- Output: `unknown`\n\n#### fetchGoogleDoc\nFetch the contents of an existing Google Document.\n- exportType controls the output format: "html" for HTML markup, "markdown" for Markdown, "json" for structured JSON, "plain" for plain text.\n- Input: `{ documentId: string, connectionId?: string, exportType: "html" | "markdown" | "json" | "plain" }`\n- Output: `{ content: string }`\n\n#### fetchGoogleSheet\nFetch contents of a Google Spreadsheet range.\n- range uses A1 notation (e.g. "Sheet1!A1:C10"). Omit to fetch the entire first sheet.\n- exportType controls the output format: "csv" for comma-separated values, "json" for structured JSON.\n- Input: `{ spreadsheetId: string, range: string, connectionId?: string, exportType: "csv" | "json" }`\n- Output: `{ content: string }`\n\n#### getGoogleSheetInfo\nGet metadata about a Google Spreadsheet including sheet names, row counts, and column counts.\n- Requires a Google OAuth connection with Drive scope.\n- Returns the spreadsheet title and a list of all sheets with their dimensions.\n- Input: `{ documentId: string, connectionId?: string }`\n- Output: `{ title: string, sheets: { sheetId: number, title: string, rowCount: number, columnCount: number }[] }`\n\n#### updateGoogleDoc\nUpdate the contents of an existing Google Document.\n- operationType controls how content is applied: "addToTop" prepends, "addToBottom" appends, "overwrite" replaces all content.\n- textType determines how the text field is interpreted: "plain" for plain text, "html" for HTML markup, "markdown" for Markdown.\n- Input: `{ documentId: string, connectionId?: string, text: string, textType: "plain" | "html" | "markdown", operationType: "addToTop" | "addToBottom" | "overwrite" }`\n- Output: `{ documentUrl: string }`\n\n#### updateGoogleSheet\nUpdate a Google Spreadsheet with new data.\n- operationType controls how data is written: "addToBottom" appends rows, "overwrite" replaces all data, "range" writes to a specific cell range.\n- Data should be provided as CSV in the text field.\n- Input: `{ text: string, connectionId?: string, spreadsheetId: string, range: string, operationType: "addToBottom" | "overwrite" | "range" }`\n- Output: `{ spreadsheetUrl: string }`\n\n### Google Calendar\n\n#### createGoogleCalendarEvent\nCreate a new event on a Google Calendar.\n- Requires a Google OAuth connection with Calendar events scope.\n- Date/time values must be ISO 8601 format (e.g. "2025-07-02T10:00:00-07:00").\n- Attendees are specified as one email address per line in a single string.\n- Set addMeetLink to true to automatically attach a Google Meet video call.\n- Input: `{ connectionId?: string, summary: string, description?: string, location?: string, startDateTime: string, endDateTime: string, attendees?: string, addMeetLink?: boolean, calendarId?: string }`\n- Output: `{ eventId: string, htmlLink: string }`\n\n#### deleteGoogleCalendarEvent\nDelete an event from a Google Calendar by event ID.\n- Requires a Google OAuth connection with Calendar events scope.\n- Throws an error if the event does not exist.\n- Input: `{ connectionId?: string, eventId: string, calendarId?: string }`\n- Output: `unknown`\n\n#### getGoogleCalendarEvent\nRetrieve a specific event from a Google Calendar by event ID.\n- Requires a Google OAuth connection with Calendar events scope.\n- The variable receives JSON or XML-like text depending on exportType. The direct execution output always returns the structured event.\n- Input: `{ connectionId?: string, eventId: string, exportType: "json" | "text", calendarId?: string }`\n- Output: `{ event: { id?: string | null, status?: string | null, htmlLink?: string | null, created?: string | null, updated?: string | null, summary?: string | null, description?: string | null, location?: string | null, organizer?: { displayName?: string | null, email?: string | null } | null, start?: { dateTime?: string | null, timeZone?: string | null } | null, end?: { dateTime?: string | null, timeZone?: string | null } | null, attendees?: ({ displayName?: string | null, email?: string | null, responseStatus?: string | null })[] | null } }`\n\n#### listGoogleCalendarEvents\nList upcoming events from a Google Calendar, ordered by start time.\n- Requires a Google OAuth connection with Calendar events scope.\n- Only returns future events (timeMin = now).\n- The variable receives JSON or XML-like text depending on exportType. The direct execution output always returns structured events.\n- Input: `{ connectionId?: string, limit: number, exportType: "json" | "text", calendarId?: string }`\n- Output: `{ events: ({ id?: string | null, status?: string | null, htmlLink?: string | null, created?: string | null, updated?: string | null, summary?: string | null, description?: string | null, location?: string | null, organizer?: { displayName?: string | null, email?: string | null } | null, start?: { dateTime?: string | null, timeZone?: string | null } | null, end?: { dateTime?: string | null, timeZone?: string | null } | null, attendees?: ({ displayName?: string | null, email?: string | null, responseStatus?: string | null })[] | null })[] }`\n\n#### searchGoogleCalendarEvents\nSearch for events in a Google Calendar by keyword, date range, or both.\n- Requires a Google OAuth connection with Calendar events scope.\n- Supports keyword search via "query" and date filtering via "timeMin"/"timeMax" (ISO 8601 format).\n- Unlike "List Events" which only shows future events, this allows searching past events too.\n- Input: `{ query?: string, timeMin?: string, timeMax?: string, calendarId?: string, limit?: number, exportType: "json" | "text", connectionId?: string }`\n- Output: `{ events: ({ id?: string | null, status?: string | null, htmlLink?: string | null, created?: string | null, updated?: string | null, summary?: string | null, description?: string | null, location?: string | null, organizer?: { displayName?: string | null, email?: string | null } | null, start?: { dateTime?: string | null, timeZone?: string | null } | null, end?: { dateTime?: string | null, timeZone?: string | null } | null, attendees?: ({ displayName?: string | null, email?: string | null, responseStatus?: string | null })[] | null })[] }`\n\n#### updateGoogleCalendarEvent\nUpdate an existing event on a Google Calendar. Only specified fields are changed.\n- Requires a Google OAuth connection with Calendar events scope.\n- Fetches the existing event first, then applies only the provided updates. Omitted fields are left unchanged.\n- Attendees are specified as one email address per line, and replace the entire attendee list.\n- Input: `{ connectionId?: string, eventId: string, summary?: string, description?: string, location?: string, startDateTime?: string, endDateTime?: string, attendees?: string, calendarId?: string }`\n- Output: `{ eventId: string, htmlLink: string }`\n\n### Google Drive\n\n#### getGoogleDriveFile\nDownload a file from Google Drive and rehost it on the CDN. Returns a public CDN URL.\n- Requires a Google OAuth connection with Drive scope.\n- Google-native files (Docs, Sheets, Slides) cannot be downloaded \u2014 use dedicated steps instead.\n- Maximum file size: 200MB.\n- The file is downloaded and stored on the public CDN, where the returned URL is publicly accessible \u2014 pass `store` to put it in one of the app\'s own file stores instead, which for a private store means the URL only loads for an authorized session.\n- Input: `{ fileId: string, connectionId?: string }`\n- Output: `{ url: string, name: string, mimeType: string, size: number }`\n\n#### listGoogleDriveFiles\nList files in a Google Drive folder.\n- Requires a Google OAuth connection with Drive scope.\n- If folderId is omitted, lists files in the root folder.\n- Returns file metadata including name, type, size, and links.\n- Input: `{ folderId?: string, limit?: number, connectionId?: string, exportType: "json" | "text" }`\n- Output: `{ files: { id: string, name: string, mimeType: string, size: string, webViewLink: string, createdTime: string, modifiedTime: string }[] }`\n\n#### searchGoogleDrive\nSearch for files in Google Drive by keyword.\n- Requires a Google OAuth connection with Drive scope.\n- Searches file content and names using Google Drive\'s fullText search.\n- Input: `{ query: string, limit?: number, connectionId?: string, exportType: "json" | "text" }`\n- Output: `{ files: { id: string, name: string, mimeType: string, size: string, webViewLink: string, createdTime: string, modifiedTime: string }[] }`\n\n### HubSpot\n\n#### hubspotCreateCompany\nCreate a new company or update an existing one in HubSpot. Matches by domain.\n- Requires a HubSpot OAuth connection (connectionId).\n- If a company with the given domain already exists, it is updated. Otherwise, a new one is created.\n- Property values are type-checked against enabledProperties before being sent to HubSpot.\n- Input: `{ connectionId?: string, company: { domain: string, name: string }, enabledProperties: ({ label: string, value: string, type: "string" | "number" | "bool" })[] }`\n- Output: `{ companyId: string }`\n\n#### hubspotCreateContact\nCreate a new contact or update an existing one in HubSpot. Matches by email address.\n- Requires a HubSpot OAuth connection (connectionId).\n- If a contact with the given email already exists, it is updated. Otherwise, a new one is created.\n- If companyDomain is provided, the contact is associated with that company (creating the company if needed).\n- Property values are type-checked against enabledProperties before being sent to HubSpot.\n- Input: `{ connectionId?: string, contact: { email: string, firstname: string, lastname: string }, enabledProperties: ({ label: string, value: string, type: "string" | "number" | "bool" })[], companyDomain: string }`\n- Output: `{ contactId: string }`\n\n#### hubspotGetCompany\nLook up a HubSpot company by domain name or company ID.\n- Requires a HubSpot OAuth connection (connectionId).\n- Returns null if the company is not found.\n- When searching by domain, performs a search query then fetches the full company record.\n- Use additionalProperties to request specific HubSpot properties beyond the defaults.\n- Input: `{ connectionId?: string, searchBy: "domain" | "id", companyDomain: string, companyId: string, additionalProperties: string[] }`\n- Output: `{ company: { id: string, properties: object, createdAt: string, updatedAt: string, archived: boolean } | null }`\n\n#### hubspotGetContact\nLook up a HubSpot contact by email address or contact ID.\n- Requires a HubSpot OAuth connection (connectionId).\n- Returns null if the contact is not found.\n- Use additionalProperties to request specific HubSpot properties beyond the defaults.\n- Input: `{ connectionId?: string, searchBy: "email" | "id", contactEmail: string, contactId: string, additionalProperties: string[] }`\n- Output: `{ contact: { id: string, properties: object, createdAt: string, updatedAt: string, archived: boolean } | null }`\n\n### Hunter.io\n\n#### hunterApiCompanyEnrichment\nLook up company information by domain using Hunter.io.\n- Returns company name, description, location, industry, size, technologies, and more.\n- If the domain input is a full URL, the hostname is automatically extracted.\n- Returns null if the company is not found.\n- Input: `{ domain: string }`\n- Output: `{ data: { name: string, domain: string, description: string | null, country: string | null, state: string | null, city: string | null, industry: string | null, employees_range: string | null, logo_url: string | null, technologies: string[] } | null }`\n\n#### hunterApiDomainSearch\nSearch for email addresses associated with a domain using Hunter.io.\n- If the domain input is a full URL, the hostname is automatically extracted.\n- Returns a list of email addresses found for the domain along with organization info.\n- Input: `{ domain: string }`\n- Output: `{ data: { domain: string, disposable: boolean, webmail: boolean, accept_all: boolean, pattern: string, organization: string, country: string | null, state: string | null, emails: ({ value: string, type: string, confidence: number, first_name: string | null, last_name: string | null, position: string | null, seniority: string | null, department: string | null, linkedin: string | null, twitter: string | null, phone_number: string | null })[], linked_domains: string[] } }`\n\n#### hunterApiEmailFinder\nFind an email address for a specific person at a domain using Hunter.io.\n- Requires a first name, last name, and domain.\n- If the domain input is a full URL, the hostname is automatically extracted.\n- Returns the most likely email address with a confidence score.\n- Input: `{ domain: string, firstName: string, lastName: string }`\n- Output: `{ data: { first_name: string, last_name: string, email: string, score: number, domain: string, accept_all: boolean, position: string | null, twitter: string | null, linkedin_url: string | null, phone_number: string | null, company: string | null, sources: { domain: string, uri: string, extracted_on: string }[] } }`\n\n#### hunterApiEmailVerification\nVerify whether an email address is valid and deliverable using Hunter.io.\n- Checks email format, MX records, SMTP server, and mailbox deliverability.\n- Returns a status ("valid", "invalid", "accept_all", "webmail", "disposable", "unknown") and a score.\n- Input: `{ email: string }`\n- Output: `{ data: { status: string, result: string, score: number, email: string, regexp: boolean, gibberish: boolean, disposable: boolean, webmail: boolean, mx_records: boolean, smtp_server: boolean, smtp_check: boolean, accept_all: boolean, block: boolean, sources: { domain: string, uri: string, extracted_on: string }[] } }`\n\n#### hunterApiPersonEnrichment\nLook up professional information about a person by their email address using Hunter.io.\n- Returns name, job title, social profiles, and company information.\n- If the person is not found, returns an object with an error message instead of throwing.\n- Input: `{ email: string }`\n- Output: `{ data: { first_name: string, last_name: string, email: string, position: string | null, seniority: string | null, department: string | null, linkedin_url: string | null, twitter: string | null, phone_number: string | null, company: { name: string, domain: string, industry: string | null } | null } | { error: string } }`\n\n### LinkedIn\n\n#### postToLinkedIn\nCreate a post on LinkedIn from the connected account.\n- Requires a LinkedIn OAuth connection (connectionId).\n- Supports text posts, image posts, video posts, document posts, and article posts.\n- Attach one media type per post: image, video, document, or article.\n- Documents support PDF, PPT, PPTX, DOC, DOCX (max 100MB, 300 pages). Displays as a slideshow carousel.\n- Articles create a link preview with optional custom title, description, and thumbnail.\n- Visibility controls who can see the post.\n- Input: `{ message: string, visibility: "PUBLIC" | "CONNECTIONS", imageUrl?: string, videoUrl?: string, documentUrl?: string, articleUrl?: string, titleText?: string, descriptionText?: string, connectionId?: string }`\n- Output: `unknown`\n\n### Meshy\n\n#### meshyAnimate\nApply a preset animation to a rigged 3D character model using Meshy.\n- Requires a rig_task_id from a previously completed Meshy rigging step.\n- Select an animation from Meshy\'s library of 600+ preset animations.\n- Only works with humanoid (bipedal) rigged characters.\n- Supports post-processing: FPS change (24/25/30/60), FBX-to-USDZ conversion, or armature extraction.\n- Animation categories: DailyActions, WalkAndRun, Fighting, Dancing, BodyMovements.\n- Input: `{ rigTaskId: string, actionId: number }`\n- Output: `{ glbUrl: string, fbxUrl?: string, objUrl?: string, usdzUrl?: string, thumbnailUrl?: string, textureUrls?: object[], animations?: { name: string, glbUrl?: string, fbxUrl?: string }[], providerTaskId?: string }`\n\n#### meshyImageTo3d\nGenerate a 3D model from one or more images using Meshy. Uses the multi-image-to-3D endpoint.\n- Accepts 1-4 image URLs. All images should depict the same object from different angles for best results.\n- By default generates with textures. Set shouldTexture to false for mesh-only output.\n- Uses should_remesh: false to preserve UV mapping integrity.\n- Input: `{ imageUrls: string[], shouldTexture?: boolean, topology?: string, targetPolycount?: number, symmetryMode?: string, poseMode?: string, textureImageUrl?: string }`\n- Output: `{ glbUrl: string, fbxUrl?: string, objUrl?: string, usdzUrl?: string, thumbnailUrl?: string, textureUrls?: object[], animations?: { name: string, glbUrl?: string, fbxUrl?: string }[], providerTaskId?: string }`\n\n#### meshyRemesh\nRemesh an existing 3D model to adjust topology, polygon count, or convert formats using Meshy.\n- Provide either an input task ID (from a previous Meshy step) or a model URL.\n- Defaults to triangle topology with 30,000 target polys.\n- Useful for reducing face count before rigging (max 300k faces for rigging).\n- Input: `{ inputTaskId?: string, modelUrl?: string, topology?: string, targetPolycount?: number, resizeHeight?: number }`\n- Output: `{ glbUrl: string, fbxUrl?: string, objUrl?: string, usdzUrl?: string, thumbnailUrl?: string, textureUrls?: object[], animations?: { name: string, glbUrl?: string, fbxUrl?: string }[], providerTaskId?: string }`\n\n#### meshyRig\nAuto-rig a humanoid 3D model and generate basic walking/running animations using Meshy.\n- Only works well with standard humanoid (bipedal) models with clearly defined limbs.\n- Prefers model_url over input_task_id for cleaner rigging input.\n- Models with more than 300,000 faces should be remeshed first.\n- Returns rigged model files and optional basic animations.\n- Input: `{ inputTaskId?: string, modelUrl?: string, heightMeters?: number }`\n- Output: `{ glbUrl: string, fbxUrl?: string, objUrl?: string, usdzUrl?: string, thumbnailUrl?: string, textureUrls?: object[], animations?: { name: string, glbUrl?: string, fbxUrl?: string }[], providerTaskId?: string }`\n\n#### meshyTextTo3d\nGenerate a 3D model preview from a text prompt using Meshy. Produces an untextured mesh (preview stage).\n- Creates a text-to-3D preview task (mesh generation only, no texture).\n- Use the Meshy Texture step to apply textures to the preview.\n- Maximum prompt length is 600 characters.\n- Input: `{ prompt: string, modelType?: string, topology?: string, targetPolycount?: number, symmetryMode?: string, poseMode?: string }`\n- Output: `{ glbUrl: string, fbxUrl?: string, objUrl?: string, usdzUrl?: string, thumbnailUrl?: string, textureUrls?: object[], animations?: { name: string, glbUrl?: string, fbxUrl?: string }[], providerTaskId?: string }`\n\n#### meshyTexture\nApply or replace textures on a 3D model using a text prompt or reference image via Meshy.\n- Provide either an input task ID (from a previous Meshy step) or a model URL.\n- Provide either a text style prompt or an image style URL to guide texturing.\n- Supports .glb, .gltf, .obj, .fbx, .stl model formats when using modelUrl.\n- By default preserves original UVs (enableOriginalUv = true).\n- Works with any model source: text-to-3D previews, image-to-3D, remeshed models, or external files.\n- Input: `{ inputTaskId?: string, modelUrl?: string, textStylePrompt?: string, imageStyleUrl?: string, enableOriginalUv?: boolean, enablePbr?: boolean }`\n- Output: `{ glbUrl: string, fbxUrl?: string, objUrl?: string, usdzUrl?: string, thumbnailUrl?: string, textureUrls?: object[], animations?: { name: string, glbUrl?: string, fbxUrl?: string }[], providerTaskId?: string }`\n\n### Notion\n\n#### notionCreatePage\nCreate a new page in Notion as a child of an existing page.\n- Requires a Notion OAuth connection (connectionId).\n- Content is provided as markdown and converted to Notion blocks (headings, paragraphs, lists, code, quotes).\n- The page is created as a child of the specified parent page (pageId).\n- Input: `{ pageId: string, content: string, title: string, connectionId?: string }`\n- Output: `{ pageId: string, pageUrl: string }`\n\n#### notionUpdatePage\nUpdate the content of an existing Notion page.\n- Requires a Notion OAuth connection (connectionId).\n- Content is provided as markdown and converted to Notion blocks.\n- "append" mode adds content to the end of the page. "overwrite" mode deletes all existing blocks first.\n- Input: `{ pageId: string, content: string, mode: "append" | "overwrite", connectionId?: string }`\n- Output: `{ pageId: string, pageUrl: string }`\n\n### QuickBooks\n\n#### quickbooksApiRequest\nCall any QuickBooks Online Accounting API route using a connected company.\n- Requires a QuickBooks OAuth connection (connectionId).\n- Route is relative to /v3/company/{realmId}/, e.g. "account", "account/93", "query", "invoice".\n- Use queryParams for query-string values such as query=select * from Account.\n- Body is JSON for POST, PUT, and PATCH.\n- Input: `{ connectionId?: string, method: string, route: string, queryParams: object, body: string }`\n- Output: `{ ok: boolean, status: number, body: unknown }`\n\n### TypeSafe\n\n#### typesafeEvaluate\nAsk TypeSafe structured questions about text or JSON and get calibrated answers. The model returns a probability, a selected option, or a score. It does not write a reply. Each question is a noul (yes or no, as a probability from 0 to 1), a choice (one option from a set you define, with a probability for each), or a score (a rating against ordered levels). Put every question in the same call. The question id is yours: answers come back under the same ids. The model defaults to Jev (jev-latest).\n- state is the content to judge: a string, or a JSON object or array (a record, a chat log, application state). Text only. Turn images, audio, and video into text first.\n- questions is a map. The key is an id you choose. It is not shown to the model. The answer comes back under the same id.\n- A noul asks a yes/no question and returns noul, a probability from 0 (no) to 1 (yes). Optional criteria.true and criteria.false say what a yes and a no mean.\n- A choice picks one option. criteria is a map of option id to a short description (or null). At most 255 options. The answer includes the winning choice, a probability for every option, and confidence.\n- A score rates the state on an ordered rubric. criteria is 2 to 10 level descriptions, low to high. The answer\'s score can land between levels, and includes a probability per level plus confidence.\n- instructions may be a string, or an object that keeps the question in one field and reference data in the others. Refer to those fields by name in backticks.\n- Send every question you need in one call. The model reads the state once and answers them together.\n- model defaults to jev-latest, the current stable release. jev-preview tracks the newest release. Pin a version such as jev-1.13.0 when a threshold was tuned against that version. The response model field is the version that answered.\n- Billed per input token. Output tokens are free.\n- Input: `{ state: string | object | unknown[], questions: object, model?: string }`\n- Output: `{ model: string, answers: object, usage: { input_tokens: number, output_tokens: number } }`\n\n### X\n\n#### postToX\nCreate a post on X (Twitter) from the connected account.\n- Requires an X OAuth connection (connectionId).\n- Maximum 280 characters of text.\n- Optionally attach up to 4 media items (images, GIFs, or videos) via mediaUrls.\n- Media URLs must be publicly accessible. The service fetches and uploads them to X.\n- Supported formats: JPEG, PNG, GIF, WEBP, MP4. Images up to 5MB, videos up to 512MB.\n- Input: `{ text: string, connectionId?: string, mediaUrls?: string[] }`\n- Output: `unknown`\n\n#### searchXPosts\nSearch recent X (Twitter) posts matching a query.\n- Searches only the past 7 days of posts.\n- Query supports X API v2 search operators (up to 512 characters).\nAvailable search operators in query:\n| Operator         | Description                                      |\n| -----------------| -------------------------------------------------|\n| from:            | Posts from a specific user (e.g., from:elonmusk) |\n| to:              | Posts sent to a specific user (e.g., to:NASA)    |\n| @                | Mentions a user (e.g., @openai)                  |\n| #                | Hashtag search (e.g., #AI)                       |\n| is:retweet       | Filters retweets                                 |\n| is:reply         | Filters replies                                  |\n| has:media        | Posts containing media (images, videos, or GIFs) |\n| has:links        | Posts containing URLs                            |\n| lang:            | Filters by language (e.g., lang:en)              |\n| -                | Excludes specific terms (e.g., -spam)            |\n| ()               | Groups terms or operators (e.g., (AI OR ML))     |\n| AND, OR, NOT     | Boolean logic for combining or excluding terms   |\nConjunction-Required Operators (must be combined with a standalone operator):\n| Operator     | Description                                    |\n| ------------ | -----------------------------------------------|\n| has:media  | Posts containing media (images, videos, or GIFs) |\n| has:links  | Posts containing URLs                            |\n| is:retweet | Filters retweets                                 |\n| is:reply   | Filters replies                                  |\nFor example, has:media alone is invalid, but #AI has:media is valid.\n- Input: `{ query: string, scope: "recent" | "all", options: { startTime?: string, endTime?: string, maxResults?: number } }`\n- Output: `{ posts: { id: string, authorId: string, dateCreated: string, text: string, stats: { retweets: number, replies: number, likes: number } }[] }`\n\n### You.com\n\n#### youDotComFinanceResearch\nAsk a financial research question using You.com Finance Research and return the sourced response.\n- Use this for financial questions such as company analysis, earnings, market research, filings, macroeconomics, and due diligence.\n- researchEffort supports deep (default) or exhaustive.\n- Finance Research returns the same response shape as Web Research, but searches a finance-optimized index.\n- Use it for cited synthesis, not raw price feeds or structured time-series exports.\n- Input: `{ input: string, researchEffort?: "deep" | "exhaustive" }`\n- Output: `{ data: object | unknown[] }`\n\n#### youDotComGetPageContent\nFetch clean Markdown, HTML, or metadata for known URLs using the You.com Contents API.\n- Use this step when you already know the URLs. Use Web Search with livecrawl when You.com should discover pages from a query.\n- A single request supports up to 10 URLs.\n- Request only the formats you need. Markdown is recommended for LLM consumption.\n- Increase crawlTimeout for JavaScript-heavy pages, up to 60 seconds.\n- Individual pages can partially fail; check each returned item before processing.\n- Input: `{ urls: string[], formats?: ("markdown" | "html" | "metadata")[], crawlTimeout?: number }`\n- Output: `{ data: object | unknown[] }`\n\n#### youDotComLiveNews\nFetch live news articles through the You.com Search API and return the full structured response.\n- Defaults freshness to day for breaking or recent news.\n- Use country and language together to monitor regional or non-English news.\n- Use livecrawl: \'news\' with livecrawlFormats: [\'markdown\'] when you need full article text.\n- Use a custom freshness range like YYYY-MM-DDtoYYYY-MM-DD for historical news windows.\n- Input: `{ query: string, freshness?: string, count?: number, country?: string, language?: string, safesearch?: "off" | "moderate" | "strict", livecrawl?: "news" | "all", livecrawlFormats?: ("markdown" | "html")[] }`\n- Output: `{ data: object | unknown[] }`\n\n#### youDotComWebResearch\nAsk a research question and return a grounded You.com Research API answer with sources.\n- Use Web Search when you need raw URLs and snippets. Use Web Research when you want a synthesized answer with citations.\n- researchEffort controls depth and latency: lite, standard, deep, or exhaustive. standard is a good default.\n- sourceControl can restrict, exclude, or boost domains, and can apply freshness or country filters.\n- includeDomains cannot be combined with excludeDomains or boostDomains.\n- outputSchema returns structured output.content and is supported by standard, deep, and exhaustive, not lite.\n- Input: `{ input: string, researchEffort?: "lite" | "standard" | "deep" | "exhaustive", sourceControl?: { includeDomains?: string[], excludeDomains?: string[], boostDomains?: string[], freshness?: string, country?: string } | string, outputSchema?: object | string }`\n- Output: `{ data: object | unknown[] }`\n\n#### youDotComWebSearch\nSearch the web and news using the You.com Search API and return the full structured response.\n- Query supports You.com search operators:\n| Operator | Description | Example |\n| -------- | ----------- | ------- |\n| site: | Search within a domain and its subdomains | site:uscourts.gov |\n| filetype: | Search for a specific file type | filetype:pdf |\n| + | Require the exact term after the operator | +GAAP |\n| - | Exclude the exact term after the operator | -prs |\n| AND | Require both expressions | guitar AND Fender |\n| OR | Match either expression | guitar OR drum |\n| NOT | Negate an expression | NOT site:uscourts.gov |\n- Use livecrawl with livecrawlFormats: [\'markdown\'] when you need full page content instead of snippets.\n- Use the Get Page Content step when you already know the URLs to fetch.\n- Use freshness for recency: day, week, month, year, or YYYY-MM-DDtoYYYY-MM-DD.\n- Use country (ISO 3166-1 alpha-2) and language (BCP 47) to focus results geographically or linguistically.\n- includeDomains cannot be combined with excludeDomains or boostDomains.\n- Input: `{ query: string, count?: number, freshness?: string, country?: string, language?: string, offset?: number, safesearch?: "off" | "moderate" | "strict", livecrawl?: "web" | "news" | "all", livecrawlFormats?: ("markdown" | "html")[], crawlTimeout?: number, includeDomains?: string[], excludeDomains?: string[], boostDomains?: string[] }`\n- Output: `{ data: object | unknown[] }`\n\n### YouTube\n\n#### fetchYoutubeCaptions\nRetrieve the captions/transcript for a YouTube video.\n- Supports multiple languages via the language parameter.\n- "text" export produces timestamped plain text; "json" export produces structured transcript data.\n- Input: `{ videoUrl: string, exportType: "text" | "json", language: string }`\n- Output: `{ transcripts: { text: string, start: number }[] }`\n\n#### fetchYoutubeChannel\nRetrieve metadata and recent videos for a YouTube channel.\n- Accepts a YouTube channel URL (e.g. https://www.youtube.com/@ChannelName or /channel/ID).\n- Returns channel info and video listings as a JSON object.\n- Input: `{ channelUrl: string }`\n- Output: `object`\n\n#### fetchYoutubeComments\nRetrieve comments for a YouTube video.\n- Paginates through comments (up to 5 pages).\n- "text" export produces markdown-formatted text; "json" export produces structured comment data.\n- Input: `{ videoUrl: string, exportType: "text" | "json", limitPages: string }`\n- Output: `{ comments: { id: string, link: string, publishedDate: string, text: string, likes: number, replies: number, author: string, authorLink: string, authorImg: string }[] }`\n\n#### fetchYoutubeVideo\nRetrieve metadata for a YouTube video (title, description, stats, channel info).\n- Returns video metadata, channel info, and engagement stats.\n- Video format data is excluded from the response.\n- Input: `{ videoUrl: string }`\n- Output: `object`\n\n#### searchYoutube\nSearch for YouTube videos by keyword.\n- Supports pagination (up to 5 pages) and country/language filters.\n- Use the filter/filterType fields for YouTube search parameter (sp) filters.\n- Input: `{ query: string, limitPages: string, filter: string, filterType: string, countryCode?: string, languageCode?: string }`\n- Output: `{ results: object }`\n\n#### searchYoutubeTrends\nRetrieve trending videos on YouTube by category and region.\n- Categories: "now" (trending now), "music", "gaming", "films".\n- Supports country and language filtering.\n- Input: `{ bp: "now" | "music" | "gaming" | "films", hl: string, gl: string }`\n- Output: `object`\n\n### Helpers\n\n#### `listModels()`\nList all available AI models across all categories.\n\nOutput:\n```typescript\n{\n  models: {\n    id: string;\n    name: string;            // Display name\n    type: "llm_chat" | "image_generation" | "video_generation" | "video_analysis" | "text_to_speech" | "vision" | "transcription";\n    maxTemperature: number;\n    maxResponseSize: number;\n    inputs: object[];        // Accepted input types\n  }[]\n}\n```\n\n#### `listModelsByType(modelType)`\nList AI models filtered by type.\n- `modelType`: `"llm_chat"` | `"image_generation"` | `"video_generation"` | `"video_analysis"` | `"text_to_speech"` | `"vision"` | `"transcription"`\n- Output: same as `listModels()`\n\n#### `listModelsSummary()`\nList all available AI models (summary). Returns only id, name, type, and tags. Suitable for display or consumption inside a model context window.\n\nOutput:\n```typescript\n{\n  models: {\n    id: string;\n    name: string;\n    type: "llm_chat" | "image_generation" | "video_generation" | "video_analysis" | "text_to_speech" | "vision" | "transcription";\n    tags: string;            // Comma-separated tags\n  }[]\n}\n```\n\n#### `listModelsSummaryByType(modelType)`\nList AI models (summary) filtered by type.\n- `modelType`: `"llm_chat"` | `"image_generation"` | `"video_generation"` | `"video_analysis"` | `"text_to_speech"` | `"vision"` | `"transcription"`\n- Output: same as `listModelsSummary()`\n\n#### `listConnectors()`\nList available OAuth connector services (Slack, Google, HubSpot, etc.) and their actions. These are third-party integrations \u2014 for most tasks, use actions directly instead.\n\nOutput:\n```typescript\n{\n  services: {\n    id: string;\n    name: string;\n    icon: string;\n    actions: { id: string; name: string }[];\n  }[]\n}\n```\n\n#### `getConnector(serviceId)`\nGet details for a single OAuth connector service by ID.\n\nOutput:\n```typescript\n{\n  service: {\n    id: string;\n    name: string;\n    icon: string;\n    actions: { id: string; name: string }[];\n  }\n}\n```\n\n#### `getConnectorAction(serviceId, actionId)`\nGet the full configuration for an OAuth connector action, including all input fields needed to call it via `runFromConnectorRegistry`. OAuth connectors are sourced from the open-source MindStudio Connector Registry (MSCR) with 850+ actions across third-party services.\n\nOutput:\n```typescript\n{\n  action: {\n    id: string;\n    name: string;\n    description: string;\n    quickHelp: string;\n    configuration: { title: string; items: { label: string; helpText: string; variable: string; type: string; defaultValue: string; placeholder: string; selectOptions?: object }[] }[];\n  }\n}\n```\n\n#### `listConnections()`\nList OAuth connections for the organization (authenticated third-party service links). Use the returned connection IDs when calling OAuth connector actions. Connectors require the user to connect to the third-party service in MindStudio before they can be used.\n\nOutput:\n```typescript\n{\n  connections: {\n    id: string;       // Connection ID to pass to connector actions\n    provider: string; // Integration provider (e.g. slack, google)\n    name: string;     // Display name or account identifier\n  }[]\n}\n```\n\n#### `estimateStepCost(stepType, step?, options?)`\nEstimate the cost of executing a step before running it. Pass the same step config you would use for execution.\n\n```typescript\nconst estimate = await mindstudio.estimateStepCost(\'generateText\', { message: \'Hello\' });\n```\n\n- `stepType`: string \u2014 The action name (e.g. `"generateText"`).\n- `step`: object \u2014 Optional action input parameters for more accurate estimates.\n- `options`: `{ appId?: string, workflowId?: string }` \u2014 Optional context for pricing.\n\nOutput:\n```typescript\n{\n  costType?: string;  // "free" when the step has no cost\n  estimates?: {\n    eventType: string;       // Billing event type\n    label: string;           // Human-readable cost label\n    unitPrice: number;       // Price per unit in nanodollars (1e-9 USD)\n    unitType: string;        // What constitutes a unit (e.g. "token", "request")\n    estimatedCost?: number;  // Estimated total cost in nanodollars, or null if not estimable\n    quantity: number;        // Number of billable units\n  }[]\n}\n```\n\n#### `changeName(displayName)`\nUpdate the display name of the authenticated agent. Useful for agents to set their own name after connecting.\n\n```typescript\nawait mindstudio.changeName(\'My Agent\');\n```\n\n#### `changeProfilePicture(profilePictureUrl)`\nUpdate the profile picture of the authenticated agent. Useful for agents to set their own avatar after connecting.\n\n```typescript\nawait mindstudio.changeProfilePicture(\'https://example.com/avatar.png\');\n```\n\n#### `uploadFile(content, options)`\n**Deprecated for app file storage** \u2014 use the `files` store below (`files.defineStore(...).put(...)`): private by default, app-scoped, and served on the app\'s own domain. `uploadFile` uploads to the shared account media CDN and remains only for account-level assets (e.g. an agent avatar for `changeProfilePicture`). Gets a signed upload URL, PUTs the file content, and returns the permanent public URL.\n\n```typescript\nimport { readFileSync } from \'fs\';\nconst { url } = await mindstudio.uploadFile(readFileSync(\'photo.png\'), { extension: \'png\', type: \'image/png\' });\n```\n\n- `content`: `Buffer | Uint8Array` \u2014 The file content.\n- `options.extension`: string \u2014 File extension without the dot (e.g. `"png"`, `"jpg"`, `"mp4"`).\n- `options.type`: string (optional) \u2014 MIME type (e.g. `"image/png"`). Determines which CDN subdomain is used.\n\nOutput: `{ url: string }` \u2014 The permanent public CDN URL.\n\n#### `files` \u2014 per-app file storage (private by default)\nTyped per-app file storage: user uploads, generated documents, images, marketing assets. Think of a store as a CDN-backed bucket the app talks to \u2014 NOT app-defined state like a `db` table. The API is *shaped* like the `db` namespace (define a store at module scope, then read/write), but its contents aren\'t modeled by code, and one store is shared across dev and prod (a file uploaded in the dev editor is already live in prod at the same URL). Creates are safe by default because keys default to unique (UUID, or a content-addressed hash); the operations to treat with care are `delete(key)` and overwriting a fixed key. Public objects are CDN-cached, and the CDN drops one when it is overwritten or deleted; its `Cache-Control`, set at put time via `cacheControl`, is how long a browser may keep a copy, at most an hour: auto-minted keys get the hour (the key is never reused), and named keys default to `public, max-age=300`, so an overwrite reaches every browser within ~5 minutes; pass `cacheControl` to tune (e.g. `\'public, max-age=60\'` for near-live config, `\'no-store\'` to revalidate every read). Files are served on the app\'s own domain: private stores (the default) are signed / session-authorized; public stores are world-readable, CDN-served, and images resize via query params (e.g. `?w=400&h=300&fit=cover`). Prefer this over the deprecated `uploadFile()` for app storage.\n\n```typescript\nimport { files } from \'@mindstudio-ai/agent\';\n\n// Define stores at module scope (access is pinned here).\nexport const Uploads = files.defineStore(\'uploads\');                  // private (default)\nexport const Assets  = files.defineStore(\'assets\', { access: \'public\' });\n\n// Store bytes and get a ready-to-use URL.\nconst file = await Uploads.put(buffer, { contentType: \'application/pdf\', filename: \'report.pdf\' });\nfile.url;              // stable on-domain URL for a logged-in user (drop into <img>/<a download>)\nawait file.shareUrl(); // absolute signed URL that works with NO session (email/embed); default 24h\n```\n\nStore methods:\n- `put(content, { key?, contentType?, filename?, contentAddressed?, cacheControl? })` \u2192 `StoredFile`. `content` is `Buffer | Uint8Array | string` \u2014 any size up to 5 GiB (large content transparently uploads direct to storage via a presigned POST, so there is no request-body cap). Omit `key` for a random UUID; `contentAddressed: true` derives an immutable `<sha256>.<ext>` key. `cacheControl` sets how long a browser may keep a public object, at most an hour (defaults: auto-minted keys an hour, named keys `public, max-age=300`).\n- `get(key)` \u2192 `Buffer` \xB7 `head(key)` \u2192 `StoredFile` \xB7 `exists(key)` \u2192 `boolean`\n- `list({ prefix?, cursor?, limit? })` \u2192 `{ files: StoredFile[]; cursor? }`\n- `delete(key)` \u2192 `void`\n- `shareUrl(key, { expiresIn? })` \u2192 signed absolute URL (private stores; no session needed)\n\n`StoredFile`: `{ store, key, access, size?, contentType?, updatedAt?, url, shareUrl() }`.\n\n`defineStore(name, options?)` options: `access: \'public\' | \'private\'` (default private), `maxSize` (bytes; cap for client-direct uploads), `contentTypes` (allowlist for client-direct uploads).\n\nUser uploads (client-direct \u2014 bytes go straight to storage, never through the app backend): a backend method mints a token, the frontend submits it with `@mindstudio-ai/interface`.\n```typescript\n// backend method\nconst token = await Uploads.createUploadToken({ contentType, maxSize: 25 * 1024 * 1024 }); // also accepts cacheControl\nreturn token;\n// frontend (@mindstudio-ai/interface)\nimport { platform } from \'@mindstudio-ai/interface\';\nconst { url } = await platform.upload(token, file, { onProgress });\n```\n\nGenerated assets: EVERY action that produces a file can put its output straight into one of these stores \u2014 pass the store handle as `store` in the OPTIONS object (second argument). There is no allowlist; if an action returns a file URL, it takes `store`. Optional: omit it and the asset goes to the shared MindStudio CDN as before. Reach for it when the asset belongs to the app \u2014 private to a user, or listed alongside its other files. Inputs are never published either way: a file you hand an action is read where it sits, and any intermediate the platform needs is written to private storage and dropped when the call finishes.\n\n```typescript\nconst { imageUrl } = await mindstudio.generateImage({ prompt: \'a red sneaker\' }, { store: Assets });\n```\n\n#### `db` \u2014 typed tables over the app\'s managed SQLite\n\nChainable, typed collection API over the app\'s managed databases. Define a table at module scope; every read/write is lazy until awaited, and reads/writes can be batched into one round trip with `db.batch()`.\n```typescript\nimport { db } from \'@mindstudio-ai/agent\';\n\nexport const Orders = db.defineTable<Order>(\'orders\');\n\n// Reads \u2014 filters compile to SQL WHERE when possible\nconst recent = await Orders.filter(o => o.status === \'paid\').sortBy(o => o.created_at).reverse().take(20);\n// Comparing to an outer-scope value? Use the bindings form so it compiles to SQL:\nconst mine = await Orders.filter((o, $) => o.requestedBy === $.uid, { uid: auth.userId }); // bindings: lifts closure var so filter compiles to SQL\n\n// Writes\nconst order = await Orders.push({ item: \'Laptop\', amount: 999, status: \'pending\' });\nawait Orders.update(order.id, { status: \'approved\' });\n```\n\nAggregation \u2014 `count()`, `sum(fn)`, `avg(fn)`, `countDistinct(fn)`, and `aggregate(spec)` compile to SQL aggregates (COUNT/TOTAL/AVG/GROUP BY); no rows are fetched, so they stay cheap at any table size. Never page a whole table into memory to compute a summary.\n```typescript\nconst top = await Answers\n  .filter((a, $) => a.surveyId === $.surveyId, { surveyId }) // bindings: compiles to SQL\n  .aggregate({\n    by: [\'questionId\', \'dimension\'],\n    select: { n: { count: true }, total: { sum: \'score\' }, avgScore: { avg: \'score\' }, respondents: { countDistinct: \'responseId\' } },\n    orderBy: \'total\', desc: true, limit: 20,\n  });\n// Array<{ questionId, dimension, n: number, total: number, avgScore: number | null, respondents: number }>\n```\nSelect terms: `{ count: true }` \xB7 `{ sum: \'col\' }` \xB7 `{ avg: \'col\' }` \xB7 `{ min: \'col\' }` \xB7 `{ max: \'col\' }` \xB7 `{ countDistinct: \'col\' }` (string column names). Omit `by` for whole-set aggregates (single object). Empty set: count/countDistinct 0, sum 0, avg/min/max null.\n\nRaw SQL escape hatch \u2014 `db.sql<T>(query, params?, { database? })` runs read-only raw SQL (joins, subqueries, window functions) against the app\'s own managed database. SELECT/WITH only; writes throw \u2014 use Table methods. Lazy and batchable. Last resort: prefer the typed API (including `aggregate()`) whenever it can express the query; raw rows may not exactly match the typed API\'s value representations.\n```typescript\nconst rows = await db.sql<{ questionId: string; n: number }>(\'SELECT questionId, COUNT(*) AS n FROM answers WHERE surveyId = ? GROUP BY questionId\', [surveyId]);\n```\n\n#### `dataSources` \u2014 searchable document corpora (RAG)\n\nDeclare a data source at module scope and import the handle. The platform owns parsing, chunking, embedding, storage and tenant isolation; every hit returns a citation.\n```typescript\nimport { dataSources } from \'@mindstudio-ai/agent\';\nexport const Policies = dataSources.defineDataSource(\'policies\');\n\n// search\nconst { results } = await Policies.search(\'what are the payment terms?\', { topK: 5 });\n// options: topK (default 5, max 50), scoreThreshold, filter, mode, maxPerDocument, highlight, rerank, hybrid\nconst context = results.map((r) => r.text).join(String.fromCharCode(10, 10));\n\n// narrow before ranking: metadata tags, filename, pages, required words\nawait Policies.search(\'termination clause\', {\n  filter: { metadata: { department: \'legal\', signedAt: { gte: 20250101, lte: 20260630 } }, phrase: \'notice period\' },\n  maxPerDocument: 2,  // stop one document monopolizing the results\n});\n\n// mode: \'hybrid\' (default) | \'semantic\' | \'lexical\'. Lexical skips the query\n// embedding \u2014 fastest, right for identifier-shaped queries (error codes, SKUs).\nawait Policies.search(\'ERR-7741X\', { mode: \'lexical\' });\n\n// add a document (background \u2014 poll documents() for status)\nawait Policies.add(buffer, { filename: \'policy.pdf\', contentType: \'application/pdf\', metadata: { department: \'legal\' } });\nconst docs = await Policies.documents();\n\n// walk a large corpus (a page at a time behind the scenes) to build your own index of it\nfor await (const doc of Policies.allDocuments()) { /* doc.id, doc.metadata, doc.status */ }\n\n// introspection \u2014 what is in the corpus, and how one document was split\nconst { chunkCount, pipeline } = await Policies.stats();  // counts + the config actually in effect\nconst chunks = await Policies.chunks(docs[0].id);       // why a document does or doesn\'t match\n\n// how many passages contain the words \u2014 exact, same filter grammar as search. Say\n// "N passages mention these words", never "N relevant results": no such total exists,\n// and search\'s hits are not this set (semantic hits need not contain the words).\nconst { chunks: mentions } = await Policies.count({ contains: \'client secret rotation\' });\n\n// what the matching passages are about: your own ids, asserted by add(), a mapper or a manifest\nconst { entities } = await Policies.entities({ filter: { entities: [\'vendor-acme\'] } });  // [{ id, chunks }], most first\n\n// the source\'s dictionary: your ids and their names, so passages are tagged wherever the text names them\nawait Policies.pushDictionary(vendors.map((v) => ({ id: v.id, names: [v.name, ...v.aliases] })));\nconst { mentions } = await Policies.link(\'Renewal terms agreed with ACME Corp.\');  // what it finds: [{ ids, text }]\n```\n\n`SearchHit`: `{ score, text, entities, citation: { documentId, filename, pageNumber, chunkIndex, headingPath, boundingBox?, url } }`. `citation.url` is a stable on-domain link to the source document for a signed-in user \u2014 drop it in an `<a href>`. A document is a cousin of a file and shares its share verbs: for a public app or another origin, `search(q, { shareCitations: 3600 })` returns every `citation.url` as an absolute signed link that needs no session, and `Policies.shareUrl(hit.citation, { expiresIn })` signs one document on demand (default 24h). `(documentId, chunkIndex)` is a chunk\'s stable identity \u2014 key an eval set on that, not on `text`.\n\nAlso on each hit: `retrievalRank`/`retrievalScore` (position before reranking \u2014 compare with the final position to see what reranking did). Three opt-in search options for debugging and display: `explain: true` adds `explain.{dense,lexical,matchedVia}` so you can see which half of hybrid found a hit; `expand: 1` adds `neighbors.{before,after}` for surrounding context; `highlight: true` adds `matches` \u2014 `{start, end}` offsets of query terms in `text`, for rendering highlights. None of them changes the results or their order.\n\nScore semantics: `score`\'s scale depends on the mode \u2014 cosine similarity (`semantic`), a rank-fusion reciprocal (`hybrid`; small numbers, not similarities), keyword-overlap weight (`lexical`) \u2014 and when reranking ran, it\'s the reranker\'s 0\u20131 relevance instead (the retrieval value stays in `retrievalScore`). So with reranking on (the default), put quality cutoffs on `score`. `scoreThreshold` cuts on the retrieval branch BEFORE fusion/reranking \u2014 a per-embedding-model cosine floor, not a relevance cutoff \u2014 so leave it unset unless you\'ve measured it on your corpus.\n\n`filter` fields (all AND together, and only narrow): `metadata` (per-key equals, any-of via an array, or a numeric range via `{gte?, lte?}`), `filename`, `documentIds`, `entities` (any of these ids), `pages: {min?, max?}`, `contains` (all words, any order), `phrase` (exact adjacent sequence). Document metadata is set at add time: scalars only, \u226416 keys, keys `[a-zA-Z0-9_-]`. Ranges are numeric only \u2014 to filter by date, store dates as sortable integers (`20250315`-style YYYYMMDD, or epoch seconds) and range on those. Re-adding the same bytes with different metadata or entities updates them in place \u2014 no re-processing.\n\nA document\'s `entities` name what it is about as the app\'s own ids (an asset tag, a person\'s row id), whether or not its text says so: set by `add(bytes, { entities })`, a mapper\'s `entities` or a job manifest line\'s, up to 256 per document, each 1\u2013128 printable characters with no spaces or commas, and carried by every passage. Metadata describes a document; entities say what it is about, and the app\'s own tables say how those things relate. `filter.entities` narrows to passages about any of the ids. `entities({ filter?, limit? })` lists which entities the matching passages are about, most first, as `{ id, chunks }`: approximate counts of passages, not documents (limit default 20, max 100). Filter on one entity to see what appears beside it.\n\nA source\'s DICTIONARY tags passages by what their text says: `pushDictionary(entries)` sends every `{ id, names }` the app has (a projection of its own tables: up to 250,000 entries, 16 names each, a million names in all) and replaces the dictionary; from then on every passage carries the ids whose names it mentions, beside its document\'s asserted ones, and `filter.entities`, hits and `entities()` see both. Matching is deterministic and model-free: case and a tag\'s separators don\'t matter (`P-101A`, `P101A`, `p 101a`), a name matches only as whole words (`P-101AB` is not `P-101A`), the longest name wins, and a name two ids claim tags both. A name needs three letters or digits (two in Chinese, Japanese or Korean). Passages indexed before a push are re-tagged in the background (a relink, nothing re-embedded; `dictionary()` shows it). Push it whole after each load of the tables it comes from: an unchanged dictionary commits nothing (`unchanged: true`). The result reports `ambiguousNames`, `mostClaimed` and names that can\'t match their own text (`unmatchable`). `link(text)` shows what the current dictionary finds in any text; use it to check a dictionary before relying on it. A mapper may not push one.\n\nSearch is deterministic for a fixed corpus and configuration \u2014 same query, same order \u2014 so regression checks are meaningful. `search()` also returns `latencyMs`.\n\n`search()` throws `index_warming` (HTTP 503) when a large corpus\'s index is being reloaded after an eviction \u2014 a minute or two, in the background. It means loading, not empty: tell the user the knowledge base is warming up and retry shortly. Small corpora reload inside the search and never raise it; sources on dedicated capacity are never evicted.\n\nOn dedicated capacity `search()` throws `index_building` (HTTP 503) instead, while a bulk load runs and while the index is built after one: the index is built once at the end of a load rather than behind every write, and the message carries "N of M vectors indexed". Handle it like `index_warming` \u2014 the knowledge base is being built, never empty; say so and retry later.\n\nOther methods: `stats()`, `documents()` (the first thousand, or `documents({ ids })` to poll what `add` returned), `documentsPage({ cursor?, limit?, order? })` (one page; `order: "newest"` is what just arrived) and `allDocuments()` (walk a corpus of any size, oldest first; to write what each document says into the app\'s own tables, give the source a READER instead), `chunks(documentId, {vectors?})`, `remove(documentId)`, `removeWhere(selector)`, `ensure(name?)` (rarely needed \u2014 `add` and `search` create-on-reference), and `DataSource.contentHash(bytes)` to check for an existing document before adding.\n\n`removeWhere({ metadata?, filename?, documentIds?, externalIdPrefix? })` removes every matching document (vectors, row and bytes) in pages of a thousand and returns `{ deleted, remaining }`. Chunk-level filter fields, `entities` and an empty selector are refused; deleting a whole source is an owner operation (`remy-admin datasources delete`), not something app code does.\n\nKeeping a corpus in sync with an S3 bucket: the owner connects the bucket once with `remy-admin datasources connect --source <slug> --bucket <b> --region <r> --prefix <p> --access-key-secret <NAME> --secret-key-secret <NAME>` (the two names are app secrets set with `remy-admin secrets set`; credentials are never written in code or pasted in chat). From then on `await Source.sync()` lists the bucket, diffs it by ETag against what was ingested, and runs a job over what changed \u2014 returning `{ job }` immediately; poll `Source.job(id)` or list `Source.jobs()`. A nightly sync is an ordinary cron interface job whose method calls `sync()`; nothing is spent when nothing changed. The job auto-approves under the connector\'s per-sync budget; a plan over it waits in `planned` for the owner to `remy-admin datasources jobs approve <id>`.\n\nWhen the file is not the document \u2014 a JSON record that should become markdown plus metadata, a JSONL object holding many articles, a kill notice that means "remove this story" \u2014 give the source a MAPPER: `datasources/<slug>.mapper.ts` exporting `defineMapper(Source, { map })` from `@mindstudio-ai/agent`, declared in mindstudio.json as `"dataSources": [{ "slug": "<slug>", "mapper": { "path": "dist/datasources/<slug>.mapper.ts" } }]`. `map(object)` gets `{ key, size, contentType, etag, lastModified, metadata, entities, bytes(), text(), json() }` and returns one of `documents([{ externalId, title, markdown, metadata?, entities?, replaces? }])`, `passthrough({ metadata?, entities? })` (ingest the raw object as-is), `skip(reason)`, or `deletes([externalId, ...])`; throwing is the object\'s error. Everything entering a mapped source is mapped \u2014 jobs, syncs and `add()` alike \u2014 so `add()` on a mapped source returns `documents` (possibly several) and throws `mapper_skipped` for an object the mapper refused. A mapper may call models (`runTask`) or `fetch`; every call is spend per object. The owner tests one with `remy-admin datasources map test --source <slug> --dev` before deploying, reads what a job skipped with `datasources jobs quarantine <id>`, and re-applies a fixed mapper with `datasources remap --source <slug>`. See the dataSources skill for the authoring loop.\n\nTo write what each document says into the app\'s own tables (the equipment a passage names, the relations it states, the people it quotes), give the source a READER: `datasources/<slug>.reader.ts` exporting `defineReader(Source, { read, removed, cleared? })` from `@mindstudio-ai/agent`, declared in mindstudio.json as `"dataSources": [{ "slug": "<slug>", "reader": { "path": "dist/datasources/<slug>.reader.ts", "timeoutMs"?: 60000, "concurrency"?: 16 } }]` (beside or instead of a mapper). The platform runs it over every document the source builds, however it arrived (`add()`, a sync or a job), with its own batching, retries, budgets and pauses: `read(document)` gets `{ id, sourceId, filename, contentType, externalId, metadata, entities, pageCount, chunkCount, createdAt, chunks() }`, where `chunks()` returns the passages as indexed, each with its entities; `removed(documentId)` runs when a document leaves the source; `cleared(sourceId)` once when the source is deleted. A document can be read more than once, so `read` replaces that document\'s rows as a set, keyed by its id, and `removed` deletes them. A throw fails that document and it is retried. The reader always runs against the LIVE release (its code and its database) and starts reading once the app publishes it; it may call models and `fetch`, and every call is spend per document. A reader may not add to, remove from or push a dictionary to its own source. The owner tests one with `remy-admin datasources reader test --source <slug> --dev`, sees its deliveries with `datasources reader status`, and reads what it hasn\'t with `datasources reread --source <slug> --unread`, a job priced before it runs. See the dataSources skill for the authoring loop.\n\nSupported: pdf, docx, pptx, xlsx, odt, rtf, epub, images, txt, md, json, csv, tsv, log, html.\n\nAdding the same bytes twice is free \u2014 documents are content-addressed, so a re-add is a no-op when this source has already processed those exact bytes under its current configuration. Only reconfiguring the source makes a re-add do work again, and that is an explicit owner-triggered migration, never a side effect of a deploy.\n\nEverything in `SearchOptions` \u2014 `mode`, `filter`, `rerank`, `hybrid`, `maxPerDocument`, `highlight` \u2014 is free to change and takes effect on the next search. How documents are chunked and embedded is configured on the corpus itself with `remy-admin datasources config` \u2014 not in code \u2014 and changing it on a populated source is rejected rather than silently rebuilding.\n\n**A data source is live and shared.** Unlike database tables there is no dev copy and no per-release isolation: adding or removing a document changes what the deployed app retrieves, immediately. Scenarios never reset a data source, and re-ingesting a large corpus costs real money.\n\n#### `analytics` \u2014 read the app\'s own traffic + event analytics\n\nThe platform auto-tracks pageviews (and custom events via the frontend\'s `analytics.track()`); the backend `analytics` namespace is the read side, so app code can correlate real traffic with its own data \u2014 lifetime views next to each blog post, "N people reading now", conversions beside the records that produced them. Backend only (hook token); reads the LIVE app\'s real traffic even from a dev sandbox. Rate limited per app (600 reads/min; the SDK auto-retries one 429). For pages composed of several reads, use `analytics.batch([spec, ...])` \u2014 up to 10 queries in one round trip, one rate token each, results in request order \u2014 instead of a parallel burst of `query()` calls competing with each other. NEVER swallow `rate_limited` errors into empty data: a page silently rendering zeros is worse than surfacing a retry state.\n\nOne general read, `query()`, shaped like the Plausible/GA4 query APIs:\n\n```typescript\nimport { analytics } from \'@mindstudio-ai/agent\';\n\n// Lifetime pageviews + unique visitors per page \u2014 top 10.\nconst top = await analytics.query({\n  metrics: [\'pageviews\', \'visitors\'],\n  dimensions: [\'path\'],\n  dateRange: \'all\',\n  limit: 10,\n});\n// top.results[0] \u2192 { dimensions: { path: \'/post/hello\' }, metrics: { pageviews: 4210, visitors: 1830 } }\n\n// Daily views for one post, full history.\nconst series = await analytics.query({\n  metrics: [\'pageviews\'],\n  dimensions: [\'time\'],\n  granularity: \'day\',\n  filters: [[\'is\', \'path\', [\'/post/hello\']]],\n  dateRange: \'all\',\n});\n```\n\nGrammar: `metrics` from `pageviews | visitors | visits | events`; `dimensions` is at most ONE entity dimension (`path`, `referrerHost`, `sourceCategory`, `country`, `city`, `deviceType`, `browser`, `os`, `language`, `visitorType`, `utmSource`\u2026`utmContent`, `eventName`) OR `\'time\'` (with `granularity: \'5m\'|\'hour\'|\'day\'|\'week\'|\'month\'` and optional IANA `timezone`); `filters` are `[op, dimension, values]` triples with ops `is | is_not | contains`; `dateRange` is a shorthand (`\'1h\'|\'24h\'|\'7d\'|\'30d\'|\'90d\'|\'all\'`) or `[startISO, endISO]`; `orderBy` (grouped only) accepts `pageviews` or `events`. With `dateRange: \'all\'`, granularity is a minimum \u2014 the server coarsens as history grows. `city` filter values are plain city names (any country) and always read the 90-day event detail.\n\n**Retention split \u2014 the one thing to get right:** queries whose filters are all `is` and touch at most one dimension read a rollup kept forever (full lifetime history, exact unique visitors). Cross-dimension intersections, `is_not`, and `contains` scan raw events retained for 90 days; the server clamps the window and reports it. Check `meta`: `source` (`rollup` vs `events`), `clamped`, `window.served`, and `metricsOmitted` \u2014 metrics a shape can\'t carry (e.g. `visits` on grouped reads) are omitted there, never returned as silent zeros. `meta.total` carries the unpaginated group count for paging.\n\nSpecials (same `dateRange`/filter-triple grammar): `analytics.live()` \u2192 `{count, countries, sparkline}` right now; `analytics.sources(opts?)` \u2192 per-session first-source ranking (UTM > referrer > direct) classified by category/vendor incl. AI assistants; `analytics.map(opts?)` \u2192 city lat/lon points; `analytics.aiSources(opts?)` \u2192 AI-assistant referrals per vendor; `analytics.crawlers.overview()/timeseries()/recent()` \u2192 bot/AI-crawler ingestion of the app\'s pages.\n\n#### `events` \u2014 realtime over named channels (publish/subscribe)\n\nPublish to named channels from any backend code (a method, a cron, a webhook handler); connected clients receive the payloads instantly over a platform-held stream \u2014 live dashboards, notifications, chat, collaborative UIs, multi-tab/device sync, with no polling and no WebSocket code. Different from `stream()`, which narrates ONE invocation to the caller currently waiting on it: events reach clients that were not part of the invocation at all. Backend only (hook token).\n\n```typescript\nimport { auth, events } from \'@mindstudio-ai/agent\';\n\n// The subscribe door is one of YOUR methods \u2014 auth checks first, then mint.\n// The grant is the ENTIRE subscribe-side authorization: whoever holds the\n// token receives those channels\' events for its lifetime.\nexport async function watchInbox() {\n  auth.requireRole(\'member\');\n  return await events.grant(`user:${auth.userId}`);  // { token, expiresAt, ttlSeconds }; ttlSeconds? 60\u20133600, default 900\n}\n\n// Anything can publish \u2014 one call, up to 500 channels:\nconst members = await RoomMembers.where({ roomId });\nconst { delivered } = await events.publish(\n  members.map((m) => `user:${m.userId}`),\n  { type: \'message\', roomId, msgId },\n);\n```\n\nThe frontend consumes with `events.connect({ getToken, onEvent, onConnect })` from `@mindstudio-ai/interface` \u2014 `getToken` calls the minting method, and the SDK handles reconnection and grant renewal.\n\n**Ephemeral client signals (cursors, typing, live strokes):** mint the grant with `publish` channels \u2014 `events.grant(channel, { publish: [channel] })` \u2014 and the frontend publishes DIRECTLY via `sub.publish(channels, data)`: no backend method runs per signal, the SDK coalesces to ~40ms batches, and failures drop (never retry). Client events cap at 8k serialized, 30 batches/sec per grant. Carry a client `seq` in the payload (delivery is unordered across batches; receivers drop stale seqs and interpolate). Durable state still commits through a method with an id-only nudge \u2014 signal ephemerally, commit through code.\n\n**A channel is an audience, and a method decides who is in it.** Never put two users\' data on one channel. Default to per-user channels with publish-time fan-out (as above) for membership-gated audiences \u2014 removing someone stops their events immediately; reserve a shared channel for genuinely broadcast content (a ticker, a live blog), where revocation waits out the grant TTL. For anonymous visitors `auth.userId` is null \u2014 key their channels on `session.visitorId` instead, or `user:null` becomes one channel shared by every anonymous user.\n\n**Events are at-most-once nudges.** Nothing is buffered while a client is disconnected and nothing replays on connect \u2014 clients refetch state in `onConnect` (subscribe for speed, reconcile for truth). `delivered: 0` means nobody is listening right now: normal, never an error. Every delivered frame carries the platform-stamped publish `id` (consumer dedupe key: `id` + `channel`); `publish` returns the same id. Payloads cap at 256k serialized chars, checked in the SDK before the network so oversize throws locally \u2014 when a publish carries data, publish before committing the write it announces; high-rate paths publish ids (`{ type, id }`) and let the client fetch. Channels are exact strings (letters, digits, `: _ - .` \u2014 no wildcards), \u2264500 per publish, \u2264100 per grant. Publishes and grants are scoped to the execution\'s environment (live/preview/dev) automatically, so a dev-session publish can never reach live users. Debug with `remy-admin events tail | publish | channels list`.\n';
+    llmsContent = '# @mindstudio-ai/agent\n\nTypeScript SDK, CLI, and MCP server for MindStudio. One API key gives you access to 200+ AI models (OpenAI, Anthropic, Google, Meta, xAI, DeepSeek, etc.) and 1,000+ actions including 850+ connector actions across third-party services from the open-source MindStudio Connector Registry (https://github.com/mindstudio-ai/mscr). No separate provider API keys required.\n\nThis file is the complete API reference. No other documentation is needed to use the SDK.\n\n## Recommended workflow\n\nThere are 150+ actions available. Do NOT try to read or load them all at once. Follow this discovery flow:\n\n1. **Identify yourself** \u2014 Call `changeName` to set your display name (use your name or whatever your user calls you). If you have a profile picture, pass its public image URL to `changeProfilePicture`. This helps users identify your requests in their logs.\n2. **Ask** \u2014 Use `mindstudio ask "your question"` (CLI) or the `ask` MCP tool for SDK guidance. It knows every action, model, and connector and returns working TypeScript code with real model IDs and config options. Examples: `mindstudio ask "generate an image with FLUX"`, `mindstudio ask "what models support vision?"`, `mindstudio ask "how do I send a Slack message?"`.\n3. **Browse** \u2014 For manual discovery, call `listActions` (MCP tool) or `mindstudio list-actions --summary` (CLI) to get a compact `{ action: description }` map of everything available (~3k tokens). Call `mindstudio info <action>` (CLI) for parameter details.\n4. **Call it** \u2014 Invoke the action with the required parameters. All actions share the same calling convention (see below).\n\nFor specific use cases:\n\n- **OAuth third-party integrations** (Slack, Google, HubSpot, etc.): These are optional OAuth connectors from the MindStudio Connector Registry \u2014 for most tasks, use actions directly instead. If you need a third-party integration: call `listConnectors()` to browse services \u2192 `getConnectorAction(serviceId, actionId)` for input fields \u2192 execute via `runFromConnectorRegistry`. Requires an OAuth connection set up in MindStudio first \u2014 call `listConnections()` to check available connections.\n- **Pre-built agents**: Call `listAgents()` to see what\'s available \u2192 `runAgent({ appId })` to execute one. **Important:** Not all agents are configured for API use. Do not try to run an agent just because it appears in the list \u2014 only run agents the user specifically asks you to run.\n- **Model selection**: Call `listModelsSummary()` or `listModelsSummaryByType("llm_chat")` to browse models, then pass the model ID as `modelOverride.model` to actions like `generateText`. Use the summary endpoints (not `listModels`) to keep token usage low.\n- **Cost estimation**: AI-powered actions (text generation, image generation, video, audio, etc.) cost money. Call `estimateStepCost(stepType, stepInput)` before running these and confirm with the user before proceeding \u2014 unless they\'ve explicitly given permission to go ahead. Non-AI actions (data lookups, OAuth connectors, etc.) are generally free.\n- **Task agents**: For multi-step tasks that need autonomous tool use (research, content creation, data enrichment), use `runTask()`. Provide a prompt, input, tools, an output contract, and a tool-use model id. The SDK runs a tool-use loop \u2014 calling the model, executing tools, feeding results back \u2014 until structured output is produced. Tools are SDK action names, your own app methods written as `{ appMethod: "save-vendor", description: "when to use it" }`, and/or inline functions written as `{ name, description, inputSchema, execute }` \u2014 the function runs in your process when the model calls it, and a thrown error goes back to the model as tool output to work around. App methods run as the user who invoked the method that started the task, with their roles, so the agent can read and write app state on their behalf. Prefer `outputSchema` for the output contract: plain JSON Schema (`type`/`properties`/`required`/`enum`/`items`; nullable via type arrays like `["string","null"]` \u2014 no `oneOf`/`$ref`/`nullable`). Output is validated every turn with automatic repair, `result.output` is fully typed from the schema (no generic argument, no `parsedSuccessfully` check), and the call throws `MindStudioError` code `task_output_schema_mismatch` if the agent cannot produce conforming output. Example: `runTask({ prompt: "Research this restaurant and save it", input: { name: "Tartine" }, tools: ["searchGoogle", "fetchUrl", { appMethod: "saveRestaurant", description: "Persist the researched restaurant." }], outputSchema: { type: "object", properties: { name: { type: "string" }, url: { type: ["string", "null"] } }, required: ["name"] }, model: yourModelId })`. The legacy `structuredOutputExample` option (an example object the model imitates) still works but is unvalidated \u2014 check `result.parsedSuccessfully` before using its output. Pass a current tool-use model id (call `listModels` or ask). Don\'t hardcode one.\n\n## Install\n\nStandalone binary (CLI/MCP, no dependencies):\n```bash\ncurl -fsSL https://msagent.ai/install.sh | bash\n```\n\nnpm (SDK + CLI):\n```bash\nnpm install @mindstudio-ai/agent\n```\n\nRequires Node.js >= 18.\n\n## CLI\n\nThe package includes a CLI for executing steps from the command line or scripts:\n\n```bash\n# Execute with named flags (kebab-case)\nmindstudio generate-image --prompt "A mountain landscape"\n\n# Execute with JSON input (JSON5-tolerant)\nmindstudio generate-image \'{prompt: "A mountain landscape"}\'\n\n# Extract a single output field\nmindstudio generate-image --prompt "A sunset" --output-key imageUrl\n\n# List all methods (compact JSON \u2014 best for LLM discovery)\nmindstudio list --summary\n\n# List all methods (human-readable table)\nmindstudio list\n\n# Show method details (params, types, output)\nmindstudio info generate-image\n\n# Run via npx without installing\nnpx @mindstudio-ai/agent generate-text --message "Hello"\n```\n\nAuth: run `mindstudio login`, set `MINDSTUDIO_API_KEY` env var, or pass `--api-key <key>`.\nMethod names are kebab-case on the CLI (camelCase also accepted). Flags are kebab-case (`--video-url` for `videoUrl`).\nUse `--output-key <key>` to extract a single field, `--no-meta` to strip $-prefixed metadata.\n\n### Authentication\n\n```bash\n# Interactive login (opens browser, saves key to ~/.mindstudio/config.json)\nmindstudio login\n\n# Check current auth status\nmindstudio whoami\n\n# Clear stored credentials\nmindstudio logout\n```\n\nAuth resolution order: `--api-key` flag > `MINDSTUDIO_API_KEY` env > `~/.mindstudio/config.json` > `CALLBACK_TOKEN` env.\n\n## MCP server\n\nThe package includes an MCP server exposing all methods as tools. Start by calling the `listSteps` tool to discover available methods.\n\n```bash\nmindstudio mcp\n```\n\nMCP client config (standalone binary \u2014 recommended):\n```json\n{\n  "mcpServers": {\n    "mindstudio": {\n      "command": "mindstudio",\n      "args": ["mcp"],\n      "env": { "MINDSTUDIO_API_KEY": "your-api-key" }\n    }\n  }\n}\n```\n\n## Setup\n\nInside MindStudio apps (auth is automatic):\n```typescript\nimport { mindstudio, db, auth, Roles, stream } from \'@mindstudio-ai/agent\';\n\nconst { imageUrl } = await mindstudio.generateImage({ prompt: \'a sunset\' });\n```\n\nExternal usage (API key required):\n```typescript\nimport { MindStudioAgent } from \'@mindstudio-ai/agent\';\nconst agent = new MindStudioAgent({ apiKey: \'your-key\' });\n```\n\nYour MindStudio API key authenticates all requests. MindStudio routes to the correct AI provider (OpenAI, Google, Anthropic, etc.) server-side \u2014 you do NOT need separate provider API keys.\n\n## Models\n\nDirect access to 200+ AI models from every major provider \u2014 all through a single API key, billed at cost with no markups.\n\nUse `listModels()` or `listModelsByType()` for full model details, or `listModelsSummary()` / `listModelsSummaryByType()` for a lightweight list (id, name, type, tags) suitable for LLM context windows. Pass a model ID to `modelOverride.model` in methods like `generateText` to select a specific model:\n\n```typescript\nconst { models } = await mindstudio.listModelsByType(\'llm_chat\');\nconst model = models.find(m => m.name.includes("Gemini"));\n\nconst { content } = await mindstudio.generateText({\n  message: \'Hello\',\n  modelOverride: {\n    model: model.id,\n    temperature: 0.7,\n    maxResponseTokens: 16000,\n  },\n});\n```\n\n## Calling convention\n\nEvery method has the signature:\n```typescript\nmindstudio.methodName(input: InputType, options?: { appId?, threadId?, onLog? }): Promise<OutputType & StepExecutionMeta>\n```\n\nThe first argument is the step-specific input object. The optional second argument controls thread/app context and debug logging.\n\n**Debug logging**: Pass `onLog` in the options to get real-time debug logs during execution (SSE streaming). Works on ALL step methods:\n```typescript\nawait mindstudio.generateImage({ prompt: \'...\' }, {\n  onLog: (event) => console.log(`[${event.tag}] ${event.value}`),\n});\n// event: { value: string, tag: string, ts: number }\n```\n\n**`stream()` \u2014 push real-time updates to the frontend:**\n```typescript\nimport { mindstudio, stream } from \'@mindstudio-ai/agent\';\n\n// String \u2192 \'token\' event (frontend receives via onToken, accumulated text)\nawait stream(\'Processing...\');\n\n// Object \u2192 \'data\' event (frontend receives via onStreamData, NOT onToken)\nawait stream({ status: \'generating\', progress: 50 });\n```\nThese are different event types \u2014 do not mix them up. `stream()` is silently ignored when no SSE connection is active (safe to include unconditionally). For long-running steps, combine `onLog` with `stream()` to push progress to the frontend.\n\n**Results are returned flat** \u2014 output fields are spread at the top level alongside metadata:\n\n```typescript\nconst { content } = await mindstudio.generateText({ message: \'Hello\' });\n\n// Full result shape for any method:\nconst result = await mindstudio.generateText({ message: `Hello` });\nresult.content;              // step-specific output field\nresult.$appId;               // string \u2014 app ID for this execution\nresult.$threadId;            // string \u2014 thread ID for this execution\nresult.$rateLimitRemaining;  // number | undefined \u2014 API calls remaining in rate limit window\nresult.$billingCost;         // number | undefined \u2014 cost in nanodollars (1e-9 USD; divide by 1e9 for dollars, e.g. 40000000 = $0.04)\nresult.$billingEvents;       // object[] | undefined \u2014 itemized billing events\n```\n\n## Thread persistence\n\nPass `$appId`/`$threadId` from a previous result to maintain conversation state, variable state, or other context across calls:\n\n```typescript\nconst r1 = await mindstudio.generateText({ message: \'My name is Alice\' });\nconst r2 = await mindstudio.generateText(\n  { message: \'What is my name?\' },\n  { threadId: r1.$threadId, appId: r1.$appId },\n);\n// r2.content => "Your name is Alice"\n```\n\n## Error handling\n\nAll errors throw `MindStudioError`:\n```typescript\nimport { MindStudioError } from \'@mindstudio-ai/agent\';\n\ntry {\n  await mindstudio.generateImage({ prompt: \'...\' });\n} catch (err) {\n  if (err instanceof MindStudioError) {\n    err.message; // Human-readable error message\n    err.code;    // Machine-readable code: "invalid_step_config", "api_error", "rate_limit_exceeded", "output_fetch_error"\n    err.status;  // HTTP status code (400, 401, 429, etc.)\n    err.details; // Raw error body from the API\n  }\n}\n```\n\n429 rate limit errors are retried automatically (configurable via `maxRetries`).\n\n## Low-level access\n\nFor action types not covered by generated methods:\n```typescript\nconst result = await mindstudio.executeStep(\'stepType\', { ...params });\n```\n\n## Batch execution\n\nExecute multiple steps in parallel in a single request. Maximum 50 steps per batch.\nIndividual step failures do not affect other steps \u2014 partial success is possible.\n\n```typescript\nconst result = await mindstudio.executeStepBatch([\n  { stepType: \'generateImage\', step: { prompt: \'a sunset\' } },\n  { stepType: \'textToSpeech\', step: { text: \'hello world\' } },\n], { appId?, threadId? });\n\n// Result:\nresult.results;          // BatchStepResult[] \u2014 same order as input\nresult.results[0].stepType;  // string\nresult.results[0].output;    // object | undefined (step output on success)\nresult.results[0].error;     // string | undefined (error message on failure)\nresult.results[0].billingCost; // number | undefined (cost in nanodollars on success)\nresult.totalBillingCost;  // number | undefined \u2014 nanodollars (divide by 1e9 for USD)\nresult.appId;            // string\nresult.threadId;         // string\n```\n\nCLI:\n```bash\nmindstudio batch \'[{"stepType":"generateImage","step":{"prompt":"a cat"}}]\'\ncat steps.json | mindstudio batch\n```\n\n## Methods\n\nAll methods below are called on a `MindStudioAgent` instance (`agent.methodName(...)`).\nInput shows the first argument object. Output shows the fields available on the returned result.\n\n### General\n\n#### addSubtitlesToVideo\nAutomatically add subtitles to a video\n- Can control style of text and animation\n- Input: `{ videoUrl: string, language: string, fontName: string, fontSize: number, fontWeight: "normal" | "bold" | "black", fontColor: "white" | "black" | "red" | "green" | "blue" | "yellow" | "orange" | "purple" | "pink" | "brown" | "gray" | "cyan" | "magenta", highlightColor: "white" | "black" | "red" | "green" | "blue" | "yellow" | "orange" | "purple" | "pink" | "brown" | "gray" | "cyan" | "magenta", strokeWidth: number, strokeColor: "black" | "white" | "red" | "green" | "blue" | "yellow" | "orange" | "purple" | "pink" | "brown" | "gray" | "cyan" | "magenta", backgroundColor: "black" | "white" | "red" | "green" | "blue" | "yellow" | "orange" | "purple" | "pink" | "brown" | "gray" | "cyan" | "magenta" | "none", backgroundOpacity: number, position: "top" | "center" | "bottom", yOffset: number, wordsPerSubtitle: number, enableAnimation: boolean, intermediateAsset?: boolean }`\n- Output: `{ videoUrl: string }`\n\n#### analyzeImage\nAnalyze an image using a vision model based on a text prompt.\n- Uses the configured vision model to generate a text analysis of the image.\n- The prompt should describe what to look for or extract from the image.\n- Pass imageUrl for a single image, or imageUrls for multiple images analyzed together in one request.\n- Most vision models (OpenAI, Grok, Gemini) accept multiple images in one request. Ideogram describe is single-image only.\n- Input: `{ prompt: string, imageUrl?: string, imageUrls?: string[], visionModelOverride?: { model: string, config?: object } | { model: string, temperature: number, maxResponseTokens: number, ignorePreamble?: boolean, userMessagePreprocessor?: { dataSource?: string, messageTemplate?: string, maxResults?: number, enabled?: boolean, shouldInherit?: boolean }, preamble?: string, multiModelEnabled?: boolean, editResponseEnabled?: boolean, promptCaching?: boolean, config?: object } }`\n- Output: `{ analysis: string }`\n\n#### analyzeVideo\nAnalyze a video using a video analysis model based on a text prompt.\n- Uses the configured video analysis model to generate a text analysis of the video.\n- The prompt should describe what to look for or extract from the video.\n- Input: `{ prompt: string, videoUrl: string, videoAnalysisModelOverride?: { model: string, config?: object } | { model: string, temperature: number, maxResponseTokens: number, ignorePreamble?: boolean, userMessagePreprocessor?: { dataSource?: string, messageTemplate?: string, maxResults?: number, enabled?: boolean, shouldInherit?: boolean }, preamble?: string, multiModelEnabled?: boolean, editResponseEnabled?: boolean, promptCaching?: boolean, config?: object } }`\n- Output: `{ analysis: string }`\n\n#### captureThumbnail\nCapture a thumbnail from a video at a specified timestamp\n- Input: `{ videoUrl: string, at: number | string }`\n- Output: `{ thumbnailUrl: string }`\n\n#### checkAppRole\nCheck whether the current user has a specific app role and branch accordingly.\n- Checks if the current user has been assigned a specific role in this app.\n- If the user has the role, transitions to the "has role" path.\n- If the user does not have the role, transitions to the "no role" path, or errors if no path is configured.\n- Role names are defined by the app creator and assigned to users via the app roles system.\n- The roleName field supports {{variables}} for dynamic role checks.\n- Input: `{ roleName: string, hasRoleStepId?: string, hasRoleWorkflowId?: string, noRoleStepId?: string, noRoleWorkflowId?: string }`\n- Output: `{ hasRole: boolean, userRoles: string[] }`\n\n#### convertPdfToImages\nConvert each page of a PDF document into a PNG image.\n- Each page is converted to a separate PNG.\n- The page images go to the public CDN by default, or into one of the app\'s own file stores when the call passes `store` \u2014 a private store keeps them private.\n- The source PDF is never published: one of the app\'s own files is read where it sits, and any other URL is staged privately for the renderer and discarded afterwards.\n- Returns an array of image URLs, one per page.\n- Input: `{ pdfUrl: string }`\n- Output: `{ imageUrls: string[] }`\n\n#### createDataSource\nLegacy v1 data sources. v2 apps should use `dataSources.defineDataSource()` from @mindstudio-ai/agent instead, which owns parsing, chunking, embedding and citations. Create a new empty vector data source for the current app.\n- Creates a new data source (vector database) associated with the current app version.\n- The data source is created empty \u2014 use the "Upload Data Source Document" block to add documents.\n- Returns the new data source ID which can be used in subsequent blocks.\n- Input: `{ name: string }`\n- Output: `unknown`\n\n#### createGmailDraft\nCreate a draft email in the connected Gmail account.\n- Requires a Google OAuth connection with Gmail compose scope.\n- The draft appears in the user\'s Gmail Drafts folder but is not sent.\n- messageType controls the body format: "plain" for plain text, "html" for raw HTML, "markdown" for auto-converted markdown.\n- Input: `{ to: string, subject: string, message: string, connectionId?: string, messageType: "plain" | "html" | "markdown" }`\n- Output: `{ draftId: string }`\n\n#### decomposeImage\nSplit an image into a base image and transparent layers.\n- Uses Seedream 5.0 Flash layer decomposition. There is no model picker.\n- An optional prompt can name the objects or groupings to separate.\n- The output is an array of hosted image URLs. PNG output keeps layer transparency.\n- Input: `{ image: string, prompt?: string, intermediateAsset?: boolean }`\n- Output: `{ imageUrls: string[] }`\n\n#### deleteDataSource\nLegacy v1 data sources. v2 apps should use `dataSources.defineDataSource()` from @mindstudio-ai/agent instead, which owns parsing, chunking, embedding and citations. Delete a vector data source from the current app.\n- Soft-deletes a data source (vector database) by marking it as deleted.\n- The Milvus partition is cleaned up asynchronously by a background cron job.\n- The data source must belong to the current app version.\n- Input: `{ dataSourceId: string }`\n- Output: `unknown`\n\n#### deleteDataSourceDocument\nLegacy v1 data sources. v2 apps should use `dataSources.defineDataSource()` from @mindstudio-ai/agent instead, which owns parsing, chunking, embedding and citations. Delete a single document from a data source.\n- Soft-deletes a document by marking it as deleted.\n- Requires both the data source ID and document ID.\n- After deletion, reloads vectors into Milvus so the data source reflects the change immediately.\n- Input: `{ dataSourceId: string, documentId: string }`\n- Output: `unknown`\n\n#### detectChanges\nDetect changes between runs by comparing current input against previously stored state. Routes execution based on whether a change occurred.\n- Persists state across runs using a global variable keyed to the step ID.\n- Two modes: "comparison" (default) uses strict string inequality; "ai" uses an LLM to determine if a meaningful change occurred.\n- First run always treats the value as "changed" since there is no previous state.\n- Each mode supports transitions to different steps/workflows for the "changed" and "unchanged" paths.\n- AI mode bills normally for the LLM call.\n- Input: `{ mode: "ai" | "comparison", input: string, prompt?: string, modelOverride?: { model: string, temperature: number, maxResponseTokens: number, ignorePreamble?: boolean, userMessagePreprocessor?: { dataSource?: string, messageTemplate?: string, maxResults?: number, enabled?: boolean, shouldInherit?: boolean }, preamble?: string, multiModelEnabled?: boolean, editResponseEnabled?: boolean, promptCaching?: boolean, config?: object }, previousValueVariable?: string, changedStepId?: string, changedWorkflowId?: string, unchangedStepId?: string, unchangedWorkflowId?: string }`\n- Output: `{ hasChanged: boolean, currentValue: string, previousValue: string, isFirstRun: boolean }`\n\n#### detectPII\nScan text for personally identifiable information using Microsoft Presidio.\n- In workflow mode, transitions to detectedStepId if PII is found, notDetectedStepId otherwise.\n- In direct execution, returns the detection results without transitioning.\n- If entities is empty, returns immediately with no detections.\n- Input: `{ input: string, language: string, entities: string[], detectedStepId?: string, notDetectedStepId?: string, outputLogVariable?: string | null }`\n- Output: `{ detected: boolean, detections: { entity_type: string, start: number, end: number, score: number }[] }`\n\n#### downloadVideo\nDownload a video file\n- Works with YouTube, TikTok, etc., by using ytdlp behind the scenes\n- Can save as mp4 or mp3\n- Input: `{ videoUrl: string, format: "mp4" | "mp3" }`\n- Output: `{ videoUrl: string }`\n\n#### enhanceImageGenerationPrompt\nGenerate or enhance an image generation prompt using a language model. Optionally generates a negative prompt.\n- Rewrites the user\'s prompt with added detail about style, lighting, colors, and composition.\n- Also useful for initial generation, it doesn\'t always need to be enhancing an existing prompt\n- When includeNegativePrompt is true, a second model call generates a negative prompt.\n- Input: `{ initialPrompt: string, includeNegativePrompt: boolean, negativePromptDestinationVariableName?: string, systemPrompt: string, modelOverride?: unknown }`\n- Output: `{ prompt: string, negativePrompt?: string }`\n\n#### enhanceVideoGenerationPrompt\nGenerate or enhance a video generation prompt using a language model. Optionally generates a negative prompt.\n- Rewrites the user\'s prompt with added detail about style, camera movement, lighting, and composition.\n- Also useful for initial generation, it doesn\'t always need to be enhancing an existing prompt\n- When includeNegativePrompt is true, a second model call generates a negative prompt.\n- Input: `{ initialPrompt: string, includeNegativePrompt: boolean, negativePromptDestinationVariableName?: string, systemPrompt: string, modelOverride?: unknown }`\n- Output: `{ prompt: string, negativePrompt?: string }`\n\n#### extractAudioFromVideo\nExtract audio MP3 from a video file\n- Input: `{ videoUrl: string }`\n- Output: `{ audioUrl: string }`\n\n#### extractText\nDownload a file from a URL and extract its text content. Supports PDFs (including scanned/image-based PDFs via OCR), plain text files, and other document formats.\n- Best suited for PDFs and raw text/document files. For web pages, use the scrapeUrl step instead.\n- Handles both text-layer PDFs and image-based/scanned PDFs (e.g. Figma/Canva exports, scanned documents). Image-based PDFs are processed with OCR automatically \u2014 there is no need to convert PDF pages to images first.\n- Accepts a single URL, a comma-separated list of URLs, or a JSON array of URLs.\n- Nothing is published. One of the app\'s own files \u2014 including a private one, or a share link \u2014 is read where it sits; any other URL is downloaded and staged privately for the extraction model, then discarded when extraction finishes.\n- Optionally set `model` to a specific document-extraction model (`mistral-ocr-latest`, `llamaparse`, `google-document-ai`); omit to use the platform default.\n- Maximum file size is 50MB per URL.\n- Input: `{ url: string | string[], model?: string }`\n- Output: `{ text: string | string[] }`\n\n#### fetchDataSourceDocument\nLegacy v1 data sources. v2 apps should use `dataSources.defineDataSource()` from @mindstudio-ai/agent instead, which owns parsing, chunking, embedding and citations. Fetch the full extracted text contents of a document in a data source.\n- Loads a document by ID and returns its full extracted text content.\n- The document must have been successfully processed (status "done").\n- Also returns document metadata (name, summary, word count).\n- Input: `{ dataSourceId: string, documentId: string }`\n- Output: `unknown`\n\n#### fetchSlackChannelHistory\nFetch recent message history from a Slack channel.\n- The user is responsible for connecting their Slack workspace and selecting the channel\n- Input: `{ connectionId?: string, channelId: string, limit?: number, startDate?: string, endDate?: string, includeImages?: boolean, includeRawMessage?: boolean }`\n- Output: `{ messages: { from: string, content: string, timestamp?: string, images?: string[], rawMessage?: { agent_session?: { agent_bot_user_ids?: string[], agent_statuses?: unknown[], date_status_processing_expire?: number, status?: string, title?: string }, app_id?: string, assistant_app_thread?: { first_user_thread_reply?: string, title?: string, title_blocks?: unknown[] }, attachments?: { actions?: unknown[], app_id?: string, app_unfurl_url?: string, author_icon?: string, author_id?: string, author_link?: string, author_name?: string, author_subname?: string, blocks?: unknown[], bot_id?: string, bot_team_id?: string, callback_id?: string, channel_id?: string, channel_name?: string, channel_team?: string, color?: string, fallback?: string, fields?: unknown[], file_id?: string, filename?: string, files?: unknown[], footer?: string, footer_icon?: string, from_url?: string, hide_border?: boolean, hide_color?: boolean, id?: number, image_bytes?: number, image_height?: number, image_url?: string, image_width?: number, indent?: boolean, is_app_unfurl?: boolean, is_file_attachment?: boolean, is_msg_unfurl?: boolean, is_reply_unfurl?: boolean, is_thread_root_unfurl?: boolean, list?: unknown, list_record?: unknown, list_record_id?: string, list_records?: unknown[], list_schema?: unknown[], list_view?: unknown, list_view_id?: string, message_blocks?: unknown[], metadata?: unknown, mimetype?: string, mrkdwn_in?: string[], msg_subtype?: string, original_url?: string, pretext?: string, preview?: unknown, service_icon?: string, service_name?: string, service_url?: string, size?: number, slack_file_id?: string, text?: string, thumb_height?: number, thumb_url?: string, thumb_width?: number, title?: string, title_link?: string, ts?: string, url?: string, video_html?: string, video_html_height?: number, video_html_width?: number, video_url?: string }[], blocks?: { accessory?: unknown, alt_text?: string, api_decoration_available?: boolean, app_collaborators?: string[], app_id?: string, author_name?: string, block_id?: string, bot_user_id?: string, button_label?: string, call?: unknown, call_id?: string, description?: unknown, developer_trace_id?: string, dispatch_action?: boolean, element?: unknown, elements?: unknown[], expand?: boolean, external_id?: string, fallback?: string, fields?: unknown[], file?: unknown, file_id?: string, function_trigger_id?: string, hint?: unknown, image_bytes?: number, image_height?: number, image_url?: string, image_width?: number, is_animated?: boolean, is_workflow_app?: boolean, label?: unknown, optional?: boolean, owning_team_id?: string, provider_icon_url?: string, provider_name?: string, sales_home_workflow_app_type?: number, share_url?: string, slack_file?: unknown, source?: string, text?: unknown, thumbnail_url?: string, title?: unknown, title_url?: string, trigger_subtype?: string, trigger_type?: string, type?: unknown, url?: string, video_url?: string, workflow_id?: string }[], bot_id?: string, bot_profile?: { app_id?: string, deleted?: boolean, icons?: unknown, id?: string, name?: string, team_id?: string, updated?: number, user_id?: string }, client_msg_id?: string, display_as_bot?: boolean, edited?: { ts?: string, user?: string }, files?: { access?: string, alt_txt?: string, app_id?: string, app_name?: string, app_provides_file_work_objects?: boolean, attachments?: unknown[], blocks?: unknown[], bot_id?: string, can_toggle_canvas_lock?: boolean, canvas_creator_id?: string, canvas_printing_enabled?: boolean, canvas_readtime?: number, canvas_template_mode?: string, cc?: unknown[], channel_actions_count?: number, channel_actions_ts?: string, channels?: string[], comments_count?: number, converted_pdf?: string, created?: number, deanimate?: string, deanimate_gif?: string, display_as_bot?: boolean, dm_mpdm_users_with_file_access?: unknown[], duration_ms?: number, edit_link?: string, edit_timestamp?: number, editable?: boolean, editor?: string, editors?: string[], external_id?: string, external_type?: string, external_url?: string, favorites?: unknown[], file_access?: string, filetype?: string, from?: unknown[], groups?: string[], has_more?: boolean, has_more_shares?: boolean, has_rich_preview?: boolean, headers?: unknown, hls?: string, hls_embed?: string, id?: string, image_exif_rotation?: number, ims?: string[], initial_comment?: unknown, inline_attachment_count?: number, is_ai_suggested?: boolean, is_channel_space?: boolean, is_external?: boolean, is_modified_by_ai?: boolean, is_public?: boolean, is_restricted_sharing_enabled?: boolean, is_starred?: boolean, last_editor?: string, last_read?: number, lines?: number, lines_more?: number, linked_channel_id?: string, list_csv_download_url?: string, list_limits?: unknown, list_metadata?: unknown, media_display_type?: string, media_progress?: unknown, mimetype?: string, mode?: string, mp4?: string, mp4_low?: string, name?: string, non_owner_editable?: boolean, num_stars?: number, org_or_workspace_access?: string, original_attachment_count?: number, original_h?: string, original_w?: string, permalink?: string, permalink_public?: string, pinned_to?: string[], pjpeg?: string, plain_text?: string, pretty_type?: string, preview?: string, preview_highlight?: string, preview_is_truncated?: boolean, preview_plain_text?: string, private_channels_with_file_access_count?: number, private_file_with_access_count?: number, public_url_shared?: boolean, quip_thread_id?: string, reactions?: unknown[], saved?: unknown, sent_to_self?: boolean, shares?: unknown, show_badge?: boolean, simplified_html?: string, size?: number, skipped_shares?: boolean, source_team?: string, subject?: string, subtype?: string, team_pref_version_history_enabled?: boolean, teams_shared_with?: unknown[], template_conversion_ts?: number, template_description?: string, template_icon?: string, template_name?: string, template_title?: string, thumb_1024?: string, thumb_1024_gif?: string, thumb_1024_h?: string, thumb_1024_w?: string, thumb_160?: string, thumb_160_gif?: string, thumb_160_h?: string, thumb_160_w?: string, thumb_360?: string, thumb_360_gif?: string, thumb_360_h?: string, thumb_360_w?: string, thumb_480?: string, thumb_480_gif?: string, thumb_480_h?: string, thumb_480_w?: string, thumb_64?: string, thumb_64_gif?: string, thumb_64_h?: string, thumb_64_w?: string, thumb_720?: string, thumb_720_gif?: string, thumb_720_h?: string, thumb_720_w?: string, thumb_80?: string, thumb_800?: string, thumb_800_gif?: string, thumb_800_h?: string, thumb_800_w?: string, thumb_80_gif?: string, thumb_80_h?: string, thumb_80_w?: string, thumb_960?: string, thumb_960_gif?: string, thumb_960_h?: string, thumb_960_w?: string, thumb_gif?: string, thumb_pdf?: string, thumb_pdf_h?: string, thumb_pdf_w?: string, thumb_tiny?: string, thumb_video?: string, thumb_video_h?: number, thumb_video_w?: number, timestamp?: number, title?: string, title_blocks?: unknown[], to?: unknown[], transcription?: unknown, update_notification?: number, updated?: number, url_private?: string, url_private_download?: string, url_static_preview?: string, user?: string, user_team?: string, username?: string, vtt?: string }[], hidden?: boolean, icons?: { emoji?: string, image_36?: string, image_48?: string, image_64?: string, image_72?: string }, inviter?: string, is_locked?: boolean, language?: { is_reliable?: boolean, locale?: string }, latest_reply?: string, metadata?: { event_payload?: unknown, event_type?: string }, parent_user_id?: string, purpose?: string, reactions?: { count?: number, name?: string, url?: string, users?: string[] }[], reply_count?: number, reply_users?: string[], reply_users_count?: number, root?: { bot_id?: string, icons?: unknown, latest_reply?: string, parent_user_id?: string, reply_count?: number, reply_users?: string[], reply_users_count?: number, subscribed?: boolean, subtype?: string, text?: string, thread_ts?: string, ts?: string, type?: string, username?: string }, subscribed?: boolean, subtype?: string, team?: string, text?: string, thread_ts?: string, topic?: string, ts?: string, type?: string, upload?: boolean, user?: string, username?: string, x_files?: string[] } }[] }`\n\n#### generate3dModel\nGenerate a 3D model using a 3D generation model.\n- Text-to-3D models use the prompt field.\n- Image-to-3D and multi-view models take image URLs through the selected model\'s configuration inputs.\n- The output is standardized as a GLB URL plus optional FBX/OBJ/USDZ, thumbnail, texture maps, and provider task ID.\n- Input: `{ prompt?: string, intermediateAsset?: boolean, threeDModelOverride?: { model: string, config?: object } }`\n- Output: `{ prompt?: string, glbUrl: string, fbxUrl?: string, objUrl?: string, usdzUrl?: string, thumbnailUrl?: string, textureUrls?: object[], providerTaskId?: string, resolvedConfig?: object }`\n\n#### generateAsset\nGenerate an HTML asset and export it as a webpage, PDF, or image\n- Agents can generate HTML documents and export as webpage, PDFs, images, or videos. They do this by using the "generatePdf" block, which defines an HTML page with variables, and then the generation process renders the page to create the output and save its URL at the specified variable.\n- The template for the HTML page is generated by a separate process, and it can only use variables that have already been defined in the workflow at the time of its execution. It has full access to handlebars to render the HTML template, including a handlebars helper to render a markdown variable string as HTML (which can be useful for creating templates that render long strings). The template can also create its own simple JavaScript to do things like format dates and strings.\n- If PDF or composited image generation are part of the workflow, assistant adds the block and leaves the "source" empty. In a separate step, assistant generates a detailed request for the developer who will write the HTML.\n- Can also auto-generate HTML from a prompt (like a generate text block to generate HTML). In these cases, create a prompt with variables in the dynamicPrompt variable describing, in detail, the document to generate\n- Can either display output directly to user (foreground mode) or save the URL of the asset to a variable (background mode)\n- Input: `{ source: string, sourceType: "html" | "markdown" | "spa" | "raw" | "dynamic" | "customInterface", outputFormat: "pdf" | "png" | "html" | "mp4" | "openGraph", pageSize: "full" | "letter" | "A4" | "custom", testData: object, options?: { pageWidthPx?: number, pageHeightPx?: number, pageOrientation?: "portrait" | "landscape", rehostMedia?: boolean, videoDurationSeconds?: number }, spaSource?: { source?: string, lastCompiledSource?: string, files?: object, paths: string[], root: string, zipUrl: string }, rawSource?: string, dynamicPrompt?: string, dynamicSourceModelOverride?: { model: string, temperature: number, maxResponseTokens: number, ignorePreamble?: boolean, userMessagePreprocessor?: { dataSource?: string, messageTemplate?: string, maxResults?: number, enabled?: boolean, shouldInherit?: boolean }, preamble?: string, multiModelEnabled?: boolean, editResponseEnabled?: boolean, promptCaching?: boolean, config?: object }, transitionControl?: "default" | "native", shareControl?: "default" | "hidden", shareImageUrl?: string, intermediateAsset?: boolean }`\n- Output: `{ url: string }`\n\n#### generateImage\nGenerate an image from a text prompt using an AI model.\n- Prompts should be descriptive but concise (roughly 3\u20136 sentences).\n- Images are hosted on the public CDN by default, or in one of the app\'s own file stores when the call passes `store` \u2014 a private store keeps the image private.\n- In foreground mode, the image is displayed to the user. In background mode, the URL is saved to a variable.\n- When generateVariants is true with numVariants > 1, multiple images are generated in parallel.\n- In direct execution, foreground mode behaves as background, and userSelect variant behavior behaves as saveAll.\n- Input: `{ prompt: string, intermediateAsset?: boolean, imageModelOverride?: { model: string, config?: object }, generateVariants?: boolean, numVariants?: number, addWatermark?: boolean }`\n- Output: `{ imageUrl: string | string[] }`\n\n#### generateLipsync\nGenerate a lip sync video from provided audio and image.\n- In foreground mode, the video is displayed to the user. In background mode, the URL is saved to a variable.\n- Input: `{ intermediateAsset?: boolean, addWatermark?: boolean, lipsyncModelOverride?: { model: string, config?: object } }`\n- Output: `unknown`\n\n#### generateMusic\nGenerate an audio file from provided instructions (text) using a music model.\n- The text field contains the instructions (prompt) for the music generation.\n- In foreground mode, the audio is displayed to the user. In background mode, the URL is saved to a variable.\n- Input: `{ text: string, intermediateAsset?: boolean, musicModelOverride?: { model: string, config?: object } }`\n- Output: `unknown`\n\n#### generateSound\nGenerate a sound effect from a text description using a sound model.\n- The text field contains the description of the sound effect to generate.\n- In foreground mode, the audio is displayed to the user. In background mode, the URL is saved to a variable.\n- Input: `{ text: string, intermediateAsset?: boolean, soundModelOverride?: { model: string, config?: object } }`\n- Output: `{ audioUrl: string }`\n\n#### generateStaticVideoFromImage\nConvert a static image to an MP4\n- Can use to create slides/intertitles/slates for video composition\n- Input: `{ imageUrl: string, duration: string }`\n- Output: `{ videoUrl: string }`\n\n#### generateText\nSend a message to an AI model and return the response, or echo a system message.\n- Source "user" sends the message to an LLM and returns the model\'s response.\n- Source "system" echoes the message content directly (no AI call).\n- Mode "background" saves the result to a variable. Mode "foreground" streams it to the user (not available in direct execution).\n- Structured output (JSON/CSV) can be enforced via structuredOutputType and structuredOutputExample.\n- When executed inside a v2 app method (managed sandbox or local dev tunnel),\nLLM token output can be streamed to the frontend in real time via an SSE\nside-channel. The frontend opts in by passing { stream: true } to the method\ninvocation via @mindstudio-ai/interface. Tokens are published to Redis\npub/sub as they arrive and forwarded as SSE events on the invoke response.\nThe method code itself is unchanged \u2014 streaming is transparent to the\ndeveloper. See Sandboxes/execute.ts and the invoke handler in V2Apps for\nthe server-side plumbing.\n- Input: `{ message: string, source?: "user" | "system", modelOverride?: { model: string, temperature: number, maxResponseTokens: number, ignorePreamble?: boolean, userMessagePreprocessor?: { dataSource?: string, messageTemplate?: string, maxResults?: number, enabled?: boolean, shouldInherit?: boolean }, preamble?: string, multiModelEnabled?: boolean, editResponseEnabled?: boolean, promptCaching?: boolean, config?: object }, structuredOutputType?: "text" | "json" | "csv", structuredOutputExample?: string, chatHistoryMode?: "include" | "exclude" }`\n- Output: `{ content: string }`\n\n#### generateVideo\nGenerate a video from a text prompt using an AI model.\n- Prompts should be descriptive but concise (roughly 3\u20136 sentences).\n- Videos are hosted on the public CDN by default, or in one of the app\'s own file stores when the call passes `store` \u2014 a private store keeps the video private, with no public copy left behind.\n- In foreground mode, the video is displayed to the user. In background mode, the URL is saved to a variable.\n- When generateVariants is true with numVariants > 1, multiple videos are generated in parallel.\n- In direct execution, foreground mode behaves as background, and userSelect variant behavior behaves as saveAll.\n- Input: `{ prompt: string, intermediateAsset?: boolean, videoModelOverride?: { model: string, config?: object }, generateVariants?: boolean, numVariants?: number, addWatermark?: boolean }`\n- Output: `{ videoUrl: string | string[] }`\n\n#### getGmailAttachments\nDownload attachments from a Gmail email and store them on the app\'s own storage.\n- Requires a Google OAuth connection with Gmail readonly scope.\n- Attachments are stored and returned as URLs \u2014 on the public CDN by default, or in one of the app\'s own file stores when the call passes `store`. Use a private store for anything confidential; a public CDN URL is readable by anyone who has it.\n- Attachments larger than 25MB are skipped.\n- Use the message ID from Search Gmail Emails, List Recent Gmail Emails, or Get Gmail Email steps.\n- Input: `{ messageId: string, connectionId?: string }`\n- Output: `unknown`\n\n#### getGmailUnreadCount\nGet the number of unread emails in the connected Gmail inbox.\n- Requires a Google OAuth connection with Gmail readonly scope.\n- Returns the unread message count for the inbox label.\n- This is a lightweight call that does not fetch any email content.\n- Input: `{ connectionId?: string }`\n- Output: `unknown`\n\n#### getMediaMetadata\nGet info about a media file\n- Input: `{ mediaUrl: string }`\n- Output: `{ metadata: string }`\n\n#### imageFaceSwap\nReplace a face in an image with a face from another image using AI.\n- Requires both a target image and a face source image.\n- Output is a PNG on the public CDN, or in one of the app\'s own file stores when the call passes `store` \u2014 a private store keeps it private.\n- Input: `{ imageUrl: string, faceImageUrl: string, engine: string }`\n- Output: `{ imageUrl: string }`\n\n#### imageRemoveWatermark\nRemove watermarks from an image using AI.\n- Output is a PNG on the public CDN, or in one of the app\'s own file stores when the call passes `store` \u2014 a private store keeps it private.\n- Input: `{ imageUrl: string, engine: string, intermediateAsset?: boolean }`\n- Output: `{ imageUrl: string }`\n\n#### insertVideoClips\nInsert b-roll clips into a base video at a timecode, optionally with an xfade transition.\n- Input: `{ baseVideoUrl: string, overlayVideos: { videoUrl: string, startTimeSec: number }[], transition?: string, transitionDuration?: number, useOverlayAudio?: boolean, intermediateAsset?: boolean }`\n- Output: `{ videoUrl: string }`\n\n#### listDataSources\nLegacy v1 data sources. v2 apps should use `dataSources.defineDataSource()` from @mindstudio-ai/agent instead, which owns parsing, chunking, embedding and citations. List all data sources for the current app.\n- Returns metadata for every data source associated with the current app version.\n- Each entry includes the data source ID, name, description, status, and document list.\n- Input: `object`\n- Output: `unknown`\n\n#### listGmailLabels\nList all labels in the connected Gmail account. Use these label IDs or names with the Update Gmail Labels step.\n- Requires a Google OAuth connection with Gmail readonly scope.\n- Returns both system labels (INBOX, SENT, TRASH, etc.) and user-created labels.\n- Label type is "system" for built-in labels or "user" for custom labels.\n- Input: `{ connectionId?: string }`\n- Output: `unknown`\n\n#### listRecentGmailEmails\nList recent emails from the connected Gmail inbox.\n- Requires a Google OAuth connection with Gmail readonly scope.\n- Returns up to 100 emails (default 5), ordered by most recent first.\n- Functionally equivalent to Search Gmail Emails with an "in:inbox" query.\n- Input: `{ connectionId?: string, exportType: "json" | "text", limit: string }`\n- Output: `unknown`\n\n#### logic\nRoute execution to different branches based on AI evaluation, comparison operators, or workflow jumps.\n- Supports two modes: "ai" (default) uses an AI model to pick the most accurate statement; "comparison" uses operator-based checks.\n- In AI mode, the model picks the most accurate statement from the list. All possible cases must be specified.\n- In comparison mode, the context is the left operand and each case\'s condition is the right operand. First matching case wins. Use operator "default" as a fallback.\n- Requires at least two cases.\n- Each case can transition to a step in the current workflow (destinationStepId) or jump to another workflow (destinationWorkflowId).\n- Input: `{ mode?: "ai" | "comparison", context: string, cases: ({ id: string, condition: string, operator?: "eq" | "neq" | "gt" | "lt" | "gte" | "lte" | "exists" | "not_exists" | "contains" | "not_contains" | "default", destinationStepId?: string, destinationWorkflowId?: string } | string)[], modelOverride?: { model: string, temperature: number, maxResponseTokens: number, ignorePreamble?: boolean, userMessagePreprocessor?: { dataSource?: string, messageTemplate?: string, maxResults?: number, enabled?: boolean, shouldInherit?: boolean }, preamble?: string, multiModelEnabled?: boolean, editResponseEnabled?: boolean, promptCaching?: boolean, config?: object } }`\n- Output: `{ selectedCase: number }`\n\n#### makeDotComRunScenario\nTrigger a Make.com (formerly Integromat) scenario via webhook and return the response.\n- The webhook URL must be configured in your Make.com scenario.\n- Input key-value pairs are sent as JSON in the POST body.\n- Response format depends on the Make.com scenario configuration.\n- Input: `{ webhookUrl: string, input: object }`\n- Output: `{ data: unknown }`\n\n#### mergeAudio\nMerge one or more clips into a single audio file.\n- Input: `{ mp3Urls: string[], fileMetadata?: object, albumArtUrl?: string, intermediateAsset?: boolean }`\n- Output: `{ audioUrl: string }`\n\n#### mergeVideos\nMerge one or more clips into a single video.\n- Input: `{ videoUrls: string[], transition?: string, transitionDuration?: number, intermediateAsset?: boolean }`\n- Output: `{ videoUrl: string }`\n\n#### mixAudioIntoVideo\nMix an audio track into a video\n- Input: `{ videoUrl: string, audioUrl: string, options: { keepVideoAudio?: boolean, audioGainDb?: number, videoGainDb?: number, loopAudio?: boolean }, intermediateAsset?: boolean }`\n- Output: `{ videoUrl: string }`\n\n#### muteVideo\nMute a video file\n- Input: `{ videoUrl: string, intermediateAsset?: boolean }`\n- Output: `{ videoUrl: string }`\n\n#### n8nRunNode\nTrigger an n8n workflow node via webhook and return the response.\n- The webhook URL must be configured in your n8n workflow.\n- Supports GET and POST methods with optional Basic authentication.\n- For GET requests, input values are sent as query parameters. For POST, they are sent as JSON body.\n- Input: `{ method: string, authentication: "none" | "basic" | "string", user: string, password: string, webhookUrl: string, input: object }`\n- Output: `{ data: unknown }`\n\n#### postToSlackChannel\nSend a message to a Slack channel via a connected bot.\n- The user is responsible for connecting their Slack workspace and selecting the channel\n- Supports both simple text messages and slack blocks messages\n- Text messages can use limited markdown (slack-only fomatting\u2014e.g., headers are just rendered as bold)\n- Input: `{ channelId: string, messageType: "string" | "blocks", message: string, connectionId?: string }`\n- Output: `unknown`\n\n#### postToZapier\nSend data to a Zapier Zap via webhook and return the response.\n- The webhook URL must be configured in the Zapier Zap settings\n- Input keys and values are sent as the JSON body of the POST request\n- The webhook response (JSON or plain text) is returned as the output\n- Input: `{ webhookUrl: string, input: object }`\n- Output: `{ data: unknown }`\n\n#### queryDataSource\nLegacy v1 data sources. v2 apps should use `dataSources.defineDataSource()` from @mindstudio-ai/agent instead, which owns parsing, chunking, embedding and citations. Search a vector data source (RAG) and return relevant document chunks.\n- Queries a vectorized data source and returns the most relevant chunks.\n- Useful for retrieval-augmented generation (RAG) workflows.\n- Input: `{ dataSourceId: string, query: string, maxResults: number }`\n- Output: `{ text: string, chunks: string[], query: string, citations: unknown[], latencyMs: number }`\n\n#### queryExternalDatabase\nExecute a SQL query against an external database connected to the workspace.\n- Requires a database connection configured in the workspace.\n- Supports PostgreSQL (including Supabase), MySQL, and MSSQL.\n- Results can be returned as JSON or CSV.\n- Input: `{ connectionId?: string, query: string, outputFormat: "json" | "csv" }`\n- Output: `{ data: unknown }`\n\n#### redactPII\nReplace personally identifiable information in text with placeholders using Microsoft Presidio.\n- PII is replaced with entity type placeholders (e.g. "Call me at <PHONE_NUMBER>").\n- If entities is empty, returns empty text immediately without processing.\n- Input: `{ input: string, language: string, entities: string[] }`\n- Output: `{ text: string }`\n\n#### removeBackgroundFromImage\nRemove the background from an image using AI, producing a transparent PNG.\n- Uses the Bria background removal model via fal.ai by default.\n- Uses WaveSpeed\'s Ideogram background removal model when type is "advanced".\n- Output is a PNG with transparency on the public CDN, or in one of the app\'s own file stores when the call passes `store` \u2014 a private store keeps it private, `autoCrop` included.\n- Input: `{ type?: "standard" | "advanced", imageUrl: string, autoCrop?: boolean }`\n- Output: `{ imageUrl: string }`\n\n#### resizeVideo\nResize a video file\n- Input: `{ videoUrl: string, mode: "fit" | "exact", maxWidth?: number, maxHeight?: number, width?: number, height?: number, strategy?: "pad" | "crop", intermediateAsset?: boolean }`\n- Output: `{ videoUrl: string }`\n\n#### runFromConnectorRegistry\nRun a raw API connector to a third-party service\n- Use the /developer/v2/helpers/connectors endpoint to list available services and actions.\n- Use /developer/v2/helpers/connectors/{serviceId}/{actionId} to get the full input configuration for an action.\n- Use /developer/v2/helpers/connections to list your available OAuth connections.\n- The actionId format is "serviceId/actionId" (e.g., "slack/send-message").\n- Pass a __connectionId to authenticate the request with a specific OAuth connection, otherwise the default will be used (if configured).\n- Input: `{ actionId: string, displayName: string, icon: string, configurationValues: object, __connectionId?: string }`\n- Output: `{ data: object }`\n\n#### scrapeLinkedInCompany\nScrape public company data from a LinkedIn company page.\n- Requires a LinkedIn company URL (e.g. https://www.linkedin.com/company/mindstudioai).\n- Returns structured company data including description, employees, updates, and similar companies.\n- Input: `{ url: string }`\n- Output: `{ company: unknown }`\n\n#### scrapeLinkedInProfile\nScrape public profile data from a LinkedIn profile page.\n- Requires a LinkedIn profile URL (e.g. https://www.linkedin.com/in/username).\n- Returns structured profile data including experience, education, articles, and activities.\n- Input: `{ url: string }`\n- Output: `{ profile: unknown }`\n\n#### scrapeUrl\nExtract text, HTML, or structured content from one or more web pages.\n- Accepts a single URL or multiple URLs (as a JSON array, comma-separated, or newline-separated).\n- Output format controls the result shape: "text" returns markdown, "html" returns raw HTML, "json" returns structured scraper data, "summary" returns a model-written summary and requires the "firecrawl" service.\n- Can optionally capture a screenshot of each page. Screenshots go to the public CDN by default, or into one of the app\'s own file stores when the call passes `store`.\n- Handles bot protection automatically; no proxy or rendering configuration is needed.\n- Input: `{ url: string, service?: "default" | "firecrawl", autoEnhance?: boolean, outputFormat?: "text" | "json" | "html" | "summary", pageOptions?: { onlyMainContent: boolean, screenshot: boolean, waitFor: number, replaceAllPathsWithAbsolutePaths: boolean, headers: object, removeTags: string[], mobile: boolean } }`\n- Output: `{ content: string | string[] | { text: string, html: string, json?: object, screenshotUrl?: string, metadata?: { title: string, description: string, url: string, image: string }, error?: { code: string, message: string }, costUnits?: number } | { text: string, html: string, json?: object, screenshotUrl?: string, metadata?: { title: string, description: string, url: string, image: string }, error?: { code: string, message: string }, costUnits?: number }[], screenshot?: string | string[] }`\n\n#### scrapeXPost\nScrape data from a single X (Twitter) post by URL.\n- Returns structured post data (text, html, optional json/screenshot/metadata).\n- Optionally saves the text content to a variable.\n- Input: `{ url: string }`\n- Output: `{ post: { text: string, html: string, json?: object, screenshotUrl?: string, metadata?: { title: string, description: string, url: string, image: string }, error?: { code: string, message: string }, costUnits?: number } }`\n\n#### scrapeXProfile\nScrape public profile data from an X (Twitter) account by URL.\n- Returns structured profile data.\n- Optionally saves the result to a variable.\n- Input: `{ url: string }`\n- Output: `{ profile: { text: string, html: string, json?: object, screenshotUrl?: string, metadata?: { title: string, description: string, url: string, image: string }, error?: { code: string, message: string }, costUnits?: number } }`\n\n#### screenshotUrl\nCapture a screenshot of a web page as a PNG image.\n- Takes a viewport or full-page screenshot of the given URL.\n- Returns a PNG image URL: on the public CDN by default, or in one of the app\'s own file stores when the call passes `store` \u2014 a private store keeps the screenshot private.\n- Viewport mode captures only the visible area; fullPage captures the entire scrollable page, scrolling it first so lazy-loaded content is included.\n- Works on sites behind bot protection (Cloudflare and similar).\n- Cookie-consent banners and similar overlays are suppressed automatically.\n- `width` and `height` are the exact output pixel dimensions.\n- You can add a delay or wait for a CSS selector before capturing. A selector that never appears fails the step rather than capturing early.\n- Input: `{ url: string, mode?: "viewport" | "fullPage", width?: number, height?: number, delay?: number, waitFor?: string }`\n- Output: `{ screenshotUrl: string }`\n\n#### searchGmailEmails\nSearch for emails in the connected Gmail account using a Gmail search query. To list recent inbox emails, pass an empty query string.\n- Requires a Google OAuth connection with Gmail readonly scope.\n- Uses Gmail search syntax (e.g. "from:user@example.com", "subject:invoice", "is:unread").\n- To list recent inbox emails, use an empty query string or "in:inbox".\n- Returns up to 100 emails (default 5). The variable receives text or JSON depending on exportType.\n- The direct execution output always returns structured email objects.\n- Input: `{ query: string, connectionId?: string, exportType: "json" | "text", limit: string }`\n- Output: `{ emails: { id: string, subject: string, from: string, to: string, date: string, plainBody: string, htmlBody: string, labels: string }[] }`\n\n#### searchGoogle\nSearch the web using Google and return structured results.\n- Defaults to us/english, but can optionally specify country and/or language.\n- Defaults to any time, but can optionally specify last hour, last day, week, month, or year.\n- Defaults to top 30 results, but can specify 1 to 100 results to return.\n- Set fetchTopN to also read the page content of the first N results in the same step, instead of following up with a separate Scrape URL step per result. Costs meaningfully more (each page is a full scrape), so ask for the fewest you need.\n- Also returns Google\'s "people also ask" and related searches when available, which are useful for choosing a follow-up query.\n- Input: `{ query: string, exportType: "text" | "json", countryCode?: string, languageCode?: string, dateRange?: "hour" | "day" | "week" | "month" | "year" | "any", numResults?: number, fetchTopN?: number }`\n- Output: `{ results: { title: string, description: string, url: string, content?: string }[], peopleAlsoAsk?: { question?: string, snippet?: string, link?: string }[], relatedSearches?: string[] }`\n\n#### searchGoogleImages\nSearch Google Images and return image results with URLs and metadata.\n- Defaults to us/english, but can optionally specify country and/or language.\n- Defaults to any time, but can optionally specify last hour, last day, week, month, or year.\n- Defaults to top 30 results, but can specify 1 to 100 results to return.\n- Input: `{ query: string, exportType: "text" | "json", countryCode?: string, languageCode?: string, dateRange?: "hour" | "day" | "week" | "month" | "year" | "any", numResults?: number }`\n- Output: `{ images: { title: string, imageUrl: string, imageWidth: number, imageHeight: number, thumbnailUrl: string, thumbnailWidth: number, thumbnailHeight: number, source: string, domain: string, link: string, googleUrl: string, position: number }[] }`\n\n#### searchGoogleNews\nSearch Google News for recent news articles matching a query.\n- Defaults to top 30 results, but can specify 1 to 100 results to return.\n- Input: `{ text: string, exportType: "text" | "json", numResults?: number }`\n- Output: `{ articles: { title: string, link: string, date: string, source: { name: string }, snippet?: string }[] }`\n\n#### searchGoogleTrends\nFetch Google Trends data for a search term.\n- date accepts shorthand ("now 1-H", "today 1-m", "today 5-y", etc.) or custom "yyyy-mm-dd yyyy-mm-dd" ranges.\n- data_type controls the shape of returned data: TIMESERIES, GEO_MAP, GEO_MAP_0, RELATED_TOPICS, or RELATED_QUERIES.\n- Input: `{ text: string, hl: string, geo: string, data_type: "TIMESERIES" | "GEO_MAP" | "GEO_MAP_0" | "RELATED_TOPICS" | "RELATED_QUERIES", cat: string, date: string, ts: string }`\n- Output: `{ trends: object }`\n\n#### searchPerplexity\nSearch the web using the Perplexity API and return structured results.\n- Defaults to US results. Use countryCode (ISO code) to filter by country.\n- Returns 10 results by default, configurable from 1 to 20.\n- The variable receives text or JSON depending on exportType. The direct execution output always returns structured results.\n- Input: `{ query: string, exportType: "text" | "json", countryCode?: string, numResults?: number }`\n- Output: `{ results: { title: string, description: string, url: string }[] }`\n\n#### sendEmail\nSend an email to one or more recipient addresses.\n- If you are writing a backend method for a v2 app, prefer mindstudio.email.send({...}) \u2014 it takes the same fields, returns the batchId you read stats back with, and sits alongside email.messages / email.batch / email.suppress. This action remains fully supported and runs the same code; it is also the only way to use connectionId or generateHtml.\n- Use the "to" field to send to one or more specific recipient email addresses directly. Allowed recipients depend on the sender: when the app sends from a domain it owns (a verified custom domain or its <slug>.madewithremy.com subdomain) any recipient is allowed; when it falls back to the shared Remy address, recipients must be verified app users or members of the app\'s organization. (v1 apps cannot use a direct "to" \u2014 they must resolve recipients via a connection.)\n- Alternatively, recipient email addresses can be resolved from OAuth connections configured by the app creator via connectionId. The user running the workflow does not specify the recipient directly.\n- Use "cc" and "bcc" to add visible / hidden recipients (a string or an array). They are subject to the same recipient rules as "to".\n- "to" is optional only in the sense that recipients can come from elsewhere \u2014 omit it and supply "cc"/"bcc" for a hidden-list send, or omit all three and recipients are resolved from an OAuth connection. Naming nobody at all is an error.\n- Bcc-only sends are supported: with no "to" or "cc", the To: header is addressed to the app\'s own sender address (the standard "undisclosed recipients" pattern) so recipients can\'t see each other. The returned "recipients" reflects that auto-filled address.\n- Every recipient counts toward the app\'s daily outbound cap, including cc and bcc.\n- The sender defaults automatically: v2 apps send from the app\'s own identity \u2014 its verified custom domain if set, else its platform subdomain (noreply@<slug>.madewithremy.com), else the default Remy address.\n- Optionally set "from" to a custom handle, but ONLY if the app has a custom domain or subdomain: a bare handle ("support" \u2192 support@<app-domain>), a full "support@your-domain.com", or "Name <support@your-domain.com>". The domain must be one the app owns, or the step fails.\n- If the body is a URL to a hosted HTML file on the CDN, the HTML is fetched and used as the email body.\n- The body is interpreted automatically: if it already looks like HTML it is sent as HTML, otherwise it is rendered from Markdown. Every email is sent as multipart with a plain-text alternative auto-derived from the body (better deliverability).\n- Set bodyType to override interpretation: "html" (send as-is), "markdown" (render to HTML), or "text" (plain text only, no HTML part). Default is "auto".\n- Optionally set "text" to supply your own plain-text alternative instead of the auto-derived one.\n- Use "attachments" to attach files by URL. Each entry is a URL string, or an object { url, filename?, contentType? } to control the attachment\'s displayed filename and MIME type.\n- When generateHtml is enabled, the body text is converted to a styled HTML email using an AI model (implies an HTML body).\n- Set replyTo to control the Reply-To address for replies.\n- For threaded replies in a shared inbox, set inReplyTo (the Message-ID being replied to) and references (prior Message-IDs in the thread).\n- connectionId can be a comma-separated list to send to multiple recipients.\n- The special connectionId "trigger_email" uses the email address that triggered the workflow.\n- Sends default to category "marketing": every message carries a one-click unsubscribe header pointing at the app\'s own domain, and recipients who previously unsubscribed from this app are skipped and returned in "suppressed". Because the unsubscribe link is per-recipient, a marketing send delivers one separate message per recipient and cc/bcc are folded into that list.\n- Set category to "transactional" for receipts, alerts, password resets and anything a recipient must not be able to opt out of. Transactional sends carry no unsubscribe header, ignore the unsubscribe list, and keep cc/bcc on a single message \u2014 use it whenever cc/bcc semantics matter.\n- The subject is stored verbatim in the app\'s delivery log, which every workspace member can read for 90 days. Never put a secret in the subject \u2014 no one-time codes, no magic-link tokens, no password-reset tokens. Put those in the body of a "transactional" send: a small transactional send is delivered immediately and its body is never stored. A "marketing" send, or any large send, is queued for delivery, which means its body is held in internal storage until the send finishes \u2014 so do not put a secret in the body of one of those either.\n- Input: `{ subject: string, body: string, to?: string | string[], cc?: string | string[], bcc?: string | string[], connectionId?: string, generateHtml?: boolean, generateHtmlInstructions?: string, generateHtmlModelOverride?: { model: string, temperature: number, maxResponseTokens: number, ignorePreamble?: boolean, userMessagePreprocessor?: { dataSource?: string, messageTemplate?: string, maxResults?: number, enabled?: boolean, shouldInherit?: boolean }, preamble?: string, multiModelEnabled?: boolean, editResponseEnabled?: boolean, promptCaching?: boolean, config?: object }, attachments?: (string | { url: string, filename?: string, contentType?: string })[], from?: string, bodyType?: "auto" | "html" | "markdown" | "text", text?: string, replyTo?: string, inReplyTo?: string, references?: string[], category?: "transactional" | "marketing", batchId?: string }`\n- Output: `{ recipients: string[], cc: string[], bcc: string[], from: string, suppressed: string[], batchId: string }`\n\n#### sendGmailDraft\nSend an existing draft from the connected Gmail account.\n- Requires a Google OAuth connection with Gmail compose scope.\n- The draft is sent and removed from the Drafts folder.\n- Use the draft ID returned by the Create Gmail Draft or List Gmail Drafts steps.\n- Input: `{ draftId: string, connectionId?: string }`\n- Output: `unknown`\n\n#### sendGmailMessage\nSend an email from the connected Gmail account.\n- Requires a Google OAuth connection with Gmail compose scope.\n- messageType controls the body format: "plain" for plain text, "html" for raw HTML, "markdown" for auto-converted markdown.\n- Input: `{ to: string, subject: string, message: string, connectionId?: string, messageType: "plain" | "html" | "markdown" }`\n- Output: `{ messageId: string }`\n\n#### sendSlackDirectMessage\nSend a direct message to a Slack user via a connected bot.\n- The user is responsible for connecting their Slack workspace\n- The recipient is identified by their Slack user ID\n- Supports both simple text messages and Slack blocks messages\n- Text messages can use limited markdown (slack-only formatting\u2014e.g., headers are just rendered as bold)\n- Input: `{ slackUserId: string, messageType: "string" | "blocks", message: string, connectionId?: string }`\n- Output: `unknown`\n\n#### sendSMS\nSend an SMS or MMS message to a phone number configured via OAuth connection.\n- User is responsible for configuring the connection to the number (MindStudio requires double opt-in to prevent spam)\n- If mediaUrls are provided, the message is sent as MMS instead of SMS\n- MMS supports up to 10 media URLs (images, video, audio, PDF) with a 5MB limit per file\n- MMS is only supported on US and Canadian carriers; international numbers will receive SMS only (media silently dropped)\n- Input: `{ body: string, connectionId?: string, mediaUrls?: string[] }`\n- Output: `unknown`\n\n#### setGmailReadStatus\nMark one or more Gmail emails as read or unread.\n- Requires a Google OAuth connection with Gmail modify scope.\n- Accepts one or more message IDs as a comma-separated string or array.\n- Set markAsRead to true to mark as read, false to mark as unread.\n- Input: `{ messageIds: string, markAsRead: boolean, connectionId?: string }`\n- Output: `unknown`\n\n#### setRunTitle\nSet the title of the agent run for the user\'s history\n- Input: `{ title: string }`\n- Output: `unknown`\n\n#### setVariable\nExplicitly set a variable to a given value.\n- Useful for bootstrapping global variables or setting constants.\n- The variable name and value both support variable interpolation.\n- The type field is a UI hint only (controls input widget in the editor).\n- Input: `{ value: string | string[] }`\n- Output: `object`\n\n#### textToSpeech\nGenerate an audio file from provided text using a speech model.\n- The text field contains the exact words to be spoken (not instructions).\n- In foreground mode, the audio is displayed to the user. In background mode, the URL is saved to a variable.\n- Input: `{ text: string, intermediateAsset?: boolean, speechModelOverride?: { model: string, config?: object } }`\n- Output: `{ audioUrl: string }`\n\n#### transcribeAudio\nConvert an audio file to text using a transcription model.\n- The prompt field provides optional context to improve transcription accuracy (e.g. language, speaker names, domain).\n- Detail beyond the plain transcript is requested through `transcriptionModelOverride.config`, and support varies by model:\n- `includeSpeakers: \'yes\'` \u2014 label each passage with a speaker. Supported by elevenlabs-scribe-v1, elevenlabs-scribe-v2, deepgram-nova-3 and grok-stt. Also switches `text` to a timestamped, speaker-labelled block format.\n- `includeWordTimings: \'yes\'` \u2014 add per-word timings and confidence inside each segment. Supported by elevenlabs-scribe-v1, elevenlabs-scribe-v2, deepgram-nova-3 and whisper-1. Off by default because an hour of speech is roughly ten thousand words.\n- `detectLanguage: \'yes\'` \u2014 detect the spoken language instead of assuming English (deepgram-nova-3 only; Scribe and the whisper models always report it).\n- The `segments` output carries timings, confidence and speakers as structured data, so there is no need to slice the audio up to recover them. It is populated for elevenlabs-scribe-v1/v2, deepgram-nova-3, whisper-1 and the DeepInfra-served whisper/voxtral/qwen models, and absent for models that return nothing but text (gpt-transcribe, gpt-4o-transcribe, grok-stt).\n- `confidence` is always 0..1 with higher meaning more confident, whichever model produced it.\n- Input: `{ audioUrl: string, prompt: string, transcriptionModelOverride?: { model: string, config?: object } }`\n- Output: `{ text: string, segments?: { start: number, end: number, text: string, confidence?: number, speaker?: string, language?: string, words?: { text: string, start?: number, end?: number, confidence?: number, speaker?: string }[] }[], language?: string, durationSeconds?: number }`\n\n#### trimMedia\nTrim an audio or video clip\n- Input: `{ inputUrl: string, start?: number | string, duration?: string | number, intermediateAsset?: boolean }`\n- Output: `{ mediaUrl: string }`\n\n#### updateGmailLabels\nAdd or remove labels on Gmail messages, identified by message IDs or a search query.\n- Requires a Google OAuth connection with Gmail modify scope.\n- Provide either a query (Gmail search syntax) or explicit messageIds to target messages.\n- Label IDs can be label names or Gmail label IDs \u2014 names are resolved automatically.\n- Input: `{ query: string, connectionId?: string, messageIds: string, addLabelIds: string, removeLabelIds: string }`\n- Output: `{ updatedMessageIds: string[] }`\n\n#### uploadDataSourceDocument\nLegacy v1 data sources. v2 apps should use `dataSources.defineDataSource()` from @mindstudio-ai/agent instead, which owns parsing, chunking, embedding and citations. Upload a file into an existing data source from a URL or raw text content.\n- If "file" is a single URL, the file is downloaded from that URL and uploaded.\n- If "file" is any other string, a .txt document is created from that content and uploaded.\n- The block waits (polls) for processing to complete before transitioning, up to 5 minutes.\n- Once processing finishes, vectors are loaded into Milvus so the data source is immediately queryable.\n- Supported file types (when using a URL) are the same as the data source upload UI (PDF, DOCX, TXT, etc.).\n- Input: `{ dataSourceId: string, file: string, fileName: string }`\n- Output: `unknown`\n\n#### upscaleImage\nIncrease the resolution of an image using AI upscaling.\n- Output is a PNG on the public CDN, or in one of the app\'s own file stores when the call passes `store` \u2014 a private store keeps it private.\n- Input: `{ imageUrl: string, targetResolution: "2k" | "4k" | "8k", engine: "standard" | "pro" }`\n- Output: `{ imageUrl: string }`\n\n#### upscaleVideo\nUpscale a video file\n- Input: `{ videoUrl: string, targetResolution: "720p" | "1080p" | "2K" | "4K", engine: "standard" | "pro" | "ultimate" | "flashvsr" | "seedance" | "seedvr2" | "runwayml/upscale-v1", intermediateAsset?: boolean }`\n- Output: `{ videoUrl: string }`\n\n#### videoFaceSwap\nSwap faces in a video file\n- Input: `{ videoUrl: string, faceImageUrl: string, targetIndex: number, engine: string, intermediateAsset?: boolean }`\n- Output: `{ videoUrl: string }`\n\n#### videoRemoveBackground\nRemove or replace background from a video\n- Input: `{ videoUrl: string, newBackground: "transparent" | "image", newBackgroundImageUrl?: string, engine: string, intermediateAsset?: boolean }`\n- Output: `{ videoUrl: string }`\n\n#### videoRemoveWatermark\nRemove a watermark from a video\n- Input: `{ videoUrl: string, engine: string, intermediateAsset?: boolean }`\n- Output: `{ videoUrl: string }`\n\n#### watermarkImage\nOverlay a watermark image onto another image.\n- The watermark is placed at the specified corner with configurable padding and width.\n- Input: `{ imageUrl: string, watermarkImageUrl: string, corner: "top-left" | "top-right" | "bottom-left" | "bottom-right", paddingPx: number, widthPx: number, intermediateAsset?: boolean }`\n- Output: `{ imageUrl: string }`\n\n#### watermarkVideo\nAdd an image watermark to a video\n- Input: `{ videoUrl: string, imageUrl: string, corner: "top-left" | "top-right" | "bottom-left" | "bottom-right", paddingPx: number, widthPx: number, intermediateAsset?: boolean }`\n- Output: `{ videoUrl: string }`\n\n### ActiveCampaign\n\n#### activeCampaignAddNote\nAdd a note to an existing contact in ActiveCampaign.\n- Requires an ActiveCampaign OAuth connection (connectionId).\n- The contact must already exist \u2014 use the contact ID from a previous create or search step.\n- Input: `{ contactId: string, note: string, connectionId?: string }`\n- Output: `unknown`\n\n#### activeCampaignCreateContact\nCreate or sync a contact in ActiveCampaign.\n- Requires an ActiveCampaign OAuth connection (connectionId).\n- If a contact with the email already exists, it may be updated depending on ActiveCampaign settings.\n- Custom fields are passed as a key-value map where keys are field IDs.\n- Input: `{ email: string, firstName: string, lastName: string, phone: string, accountId: string, customFields: object, connectionId?: string }`\n- Output: `{ contactId: string }`\n\n### Airtable\n\n#### airtableCreateUpdateRecord\nCreate a new record or update an existing record in an Airtable table.\n- If recordId is provided, updates that record. Otherwise, creates a new one.\n- When updating with updateMode "onlySpecified", unspecified fields are left as-is. With "all", unspecified fields are cleared.\n- Array fields (e.g. multipleAttachments) accept arrays of values.\n- Input: `{ connectionId?: string, baseId: string, tableId: string, recordId?: string, updateMode?: "onlySpecified" | "all", fields: unknown, recordData: object }`\n- Output: `{ recordId: string }`\n\n#### airtableDeleteRecord\nDelete a record from an Airtable table by its record ID.\n- Requires an active Airtable OAuth connection (connectionId).\n- Silently succeeds if the record does not exist.\n- Input: `{ connectionId?: string, baseId: string, tableId: string, recordId: string }`\n- Output: `{ deleted: boolean }`\n\n#### airtableGetRecord\nFetch a single record from an Airtable table by its record ID.\n- Requires an active Airtable OAuth connection (connectionId).\n- If the record is not found, returns a string message instead of a record object.\n- Input: `{ connectionId?: string, baseId: string, tableId: string, recordId: string }`\n- Output: `{ record: { id: string, createdTime: string, fields: object } | null }`\n\n#### airtableGetTableRecords\nFetch multiple records from an Airtable table with optional pagination.\n- Requires an active Airtable OAuth connection (connectionId).\n- Default limit is 100 records. Maximum is 1000.\n- When outputFormat is \'csv\', the variable receives CSV text. The direct execution output always returns parsed records.\n- Input: `{ connectionId?: string, baseId: string, tableId: string, outputFormat?: "json" | "csv", limit?: number }`\n- Output: `{ records: { id: string, createdTime: string, fields: object }[] }`\n\n### Apollo\n\n#### enrichPerson\nLook up professional information about a person using Apollo.io. Search by ID, name, LinkedIn URL, email, or domain.\n- At least one search parameter must be provided.\n- Returns enriched data from Apollo including contact details, employment info, and social profiles.\n- Input: `{ params: { id: string, name: string, linkedinUrl: string, email: string, domain: string } }`\n- Output: `{ data: unknown }`\n\n#### peopleSearch\nSearch for people matching specific criteria using Apollo.io. Supports natural language queries and advanced filters.\n- Can use a natural language "smartQuery" which is converted to Apollo search parameters by an AI model.\n- Advanced params can override or supplement the smart query results.\n- Optionally enriches returned people and/or their organizations for additional detail.\n- Results are paginated. Use limit and page to control the result window.\n- Input: `{ smartQuery: string, enrichPeople: boolean, enrichOrganizations: boolean, limit: string, page: string, params: { personTitles: string, includeSimilarTitles: string, qKeywords: string, personLocations: string, personSeniorities: string, organizationLocations: string, qOrganizationDomainsList: string, contactEmailStatus: string, organizationNumEmployeesRanges: string, revenueRangeMin: string, revenueRangeMax: string, currentlyUsingAllOfTechnologyUids: string, currentlyUsingAnyOfTechnologyUids: string, currentlyNotUsingAnyOfTechnologyUids: string } }`\n- Output: `{ results: unknown }`\n\n### Coda\n\n#### codaCreateUpdatePage\nCreate a new page or update an existing page in a Coda document.\n- Requires a Coda OAuth connection (connectionId).\n- If pageData.pageId is provided, updates that page. Otherwise, creates a new one.\n- Page content is provided as markdown and converted to Coda\'s canvas format.\n- When updating, insertionMode controls how content is applied (default: \'append\').\n- Input: `{ connectionId?: string, pageData: { docId: string, pageId?: string, name: string, subtitle: string, iconName: string, imageUrl: string, parentPageId?: string, pageContent: string | unknown, contentUpdate?: unknown, insertionMode?: string } }`\n- Output: `{ pageId: string }`\n\n#### codaCreateUpdateRow\nCreate a new row or update an existing row in a Coda table.\n- Requires a Coda OAuth connection (connectionId).\n- If rowId is provided, updates that row. Otherwise, creates a new one.\n- Row data keys are column IDs. Empty values are excluded.\n- Input: `{ connectionId?: string, docId: string, tableId: string, rowId?: string, rowData: object }`\n- Output: `{ rowId: string }`\n\n#### codaFindRow\nSearch for a row in a Coda table by matching column values.\n- Requires a Coda OAuth connection (connectionId).\n- Returns the first row matching all specified column values, or null if no match.\n- Search criteria in rowData are ANDed together.\n- Input: `{ connectionId?: string, docId: string, tableId: string, rowData: object }`\n- Output: `{ row: { id: string, values: object } | null }`\n\n#### codaGetPage\nExport and read the contents of a page from a Coda document.\n- Requires a Coda OAuth connection (connectionId).\n- Page export is asynchronous on Coda\'s side \u2014 there may be a brief delay while it processes.\n- If a page was just created in a prior step, there is an automatic 20-second retry if the first export attempt fails.\n- Input: `{ connectionId?: string, docId: string, pageId: string, outputFormat?: "html" | "markdown" }`\n- Output: `{ content: string }`\n\n#### codaGetTableRows\nFetch rows from a Coda table with optional pagination.\n- Requires a Coda OAuth connection (connectionId).\n- Default limit is 10000 rows. Rows are fetched in pages of 500.\n- When outputFormat is \'csv\', the variable receives CSV text. The direct execution output always returns parsed rows.\n- Input: `{ connectionId?: string, docId: string, tableId: string, limit?: number | string, outputFormat?: "json" | "csv" }`\n- Output: `{ rows: { id: string, values: object }[] }`\n\n### Gmail\n\n#### deleteGmailEmail\nMove an email to trash in the connected Gmail account (recoverable delete).\n- Requires a Google OAuth connection with Gmail modify scope.\n- Uses trash (recoverable) rather than permanent delete.\n- Input: `{ messageId: string, connectionId?: string }`\n- Output: `unknown`\n\n#### getGmailDraft\nRetrieve a specific draft from Gmail by draft ID.\n- Requires a Google OAuth connection with Gmail readonly scope.\n- Returns the draft content including subject, recipients, sender, and body.\n- Input: `{ draftId: string, connectionId?: string }`\n- Output: `{ draftId: string, messageId: string, subject: string, to: string, from: string, body: string }`\n\n#### getGmailEmail\nRetrieve a specific email from Gmail by message ID.\n- Requires a Google OAuth connection with Gmail readonly scope.\n- Returns the email subject, sender, recipient, date, body (plain text preferred, falls back to HTML), and labels.\n- Input: `{ messageId: string, connectionId?: string }`\n- Output: `{ messageId: string, subject: string, from: string, to: string, date: string, body: string, labels: string }`\n\n#### listGmailDrafts\nList drafts in the connected Gmail account.\n- Requires a Google OAuth connection with Gmail readonly scope.\n- Returns up to 50 drafts (default 10).\n- The variable receives text or JSON depending on exportType.\n- Input: `{ connectionId?: string, limit?: string, exportType: "json" | "text" }`\n- Output: `{ drafts: { draftId: string, messageId: string, subject: string, to: string, snippet: string }[] }`\n\n#### replyToGmailEmail\nReply to an existing email in Gmail. The reply is threaded under the original message.\n- Requires a Google OAuth connection with Gmail compose and readonly scopes.\n- The reply is sent to the original sender and threaded under the original message.\n- messageType controls the body format: "plain", "html", or "markdown".\n- Input: `{ messageId: string, message: string, messageType: "plain" | "html" | "markdown", connectionId?: string }`\n- Output: `{ messageId: string }`\n\n### Google\n\n#### createGoogleDoc\nCreate a new Google Document and optionally populate it with content.\n- textType determines how the text field is interpreted: "plain" for plain text, "html" for HTML markup, "markdown" for Markdown.\n- Input: `{ title: string, text: string, connectionId?: string, textType: "plain" | "html" | "markdown" }`\n- Output: `{ documentUrl: string }`\n\n#### createGoogleSheet\nCreate a new Google Spreadsheet and populate it with CSV data.\n- Input: `{ title: string, text: string, connectionId?: string }`\n- Output: `{ spreadsheetUrl: string }`\n\n#### deleteGoogleSheetRows\nDelete a range of rows from a Google Spreadsheet.\n- Requires a Google OAuth connection with Drive scope.\n- startRow and endRow are 1-based row numbers (inclusive).\n- If sheetName is omitted, operates on the first sheet.\n- Input: `{ documentId: string, sheetName?: string, startRow: string, endRow: string, connectionId?: string }`\n- Output: `unknown`\n\n#### fetchGoogleDoc\nFetch the contents of an existing Google Document.\n- exportType controls the output format: "html" for HTML markup, "markdown" for Markdown, "json" for structured JSON, "plain" for plain text.\n- Input: `{ documentId: string, connectionId?: string, exportType: "html" | "markdown" | "json" | "plain" }`\n- Output: `{ content: string }`\n\n#### fetchGoogleSheet\nFetch contents of a Google Spreadsheet range.\n- range uses A1 notation (e.g. "Sheet1!A1:C10"). Omit to fetch the entire first sheet.\n- exportType controls the output format: "csv" for comma-separated values, "json" for structured JSON.\n- Input: `{ spreadsheetId: string, range: string, connectionId?: string, exportType: "csv" | "json" }`\n- Output: `{ content: string }`\n\n#### getGoogleSheetInfo\nGet metadata about a Google Spreadsheet including sheet names, row counts, and column counts.\n- Requires a Google OAuth connection with Drive scope.\n- Returns the spreadsheet title and a list of all sheets with their dimensions.\n- Input: `{ documentId: string, connectionId?: string }`\n- Output: `{ title: string, sheets: { sheetId: number, title: string, rowCount: number, columnCount: number }[] }`\n\n#### updateGoogleDoc\nUpdate the contents of an existing Google Document.\n- operationType controls how content is applied: "addToTop" prepends, "addToBottom" appends, "overwrite" replaces all content.\n- textType determines how the text field is interpreted: "plain" for plain text, "html" for HTML markup, "markdown" for Markdown.\n- Input: `{ documentId: string, connectionId?: string, text: string, textType: "plain" | "html" | "markdown", operationType: "addToTop" | "addToBottom" | "overwrite" }`\n- Output: `{ documentUrl: string }`\n\n#### updateGoogleSheet\nUpdate a Google Spreadsheet with new data.\n- operationType controls how data is written: "addToBottom" appends rows, "overwrite" replaces all data, "range" writes to a specific cell range.\n- Data should be provided as CSV in the text field.\n- Input: `{ text: string, connectionId?: string, spreadsheetId: string, range: string, operationType: "addToBottom" | "overwrite" | "range" }`\n- Output: `{ spreadsheetUrl: string }`\n\n### Google Calendar\n\n#### createGoogleCalendarEvent\nCreate a new event on a Google Calendar.\n- Requires a Google OAuth connection with Calendar events scope.\n- Date/time values must be ISO 8601 format (e.g. "2025-07-02T10:00:00-07:00").\n- Attendees are specified as one email address per line in a single string.\n- Set addMeetLink to true to automatically attach a Google Meet video call.\n- Input: `{ connectionId?: string, summary: string, description?: string, location?: string, startDateTime: string, endDateTime: string, attendees?: string, addMeetLink?: boolean, calendarId?: string }`\n- Output: `{ eventId: string, htmlLink: string }`\n\n#### deleteGoogleCalendarEvent\nDelete an event from a Google Calendar by event ID.\n- Requires a Google OAuth connection with Calendar events scope.\n- Throws an error if the event does not exist.\n- Input: `{ connectionId?: string, eventId: string, calendarId?: string }`\n- Output: `unknown`\n\n#### getGoogleCalendarEvent\nRetrieve a specific event from a Google Calendar by event ID.\n- Requires a Google OAuth connection with Calendar events scope.\n- The variable receives JSON or XML-like text depending on exportType. The direct execution output always returns the structured event.\n- Input: `{ connectionId?: string, eventId: string, exportType: "json" | "text", calendarId?: string }`\n- Output: `{ event: { id?: string | null, status?: string | null, htmlLink?: string | null, created?: string | null, updated?: string | null, summary?: string | null, description?: string | null, location?: string | null, organizer?: { displayName?: string | null, email?: string | null } | null, start?: { dateTime?: string | null, timeZone?: string | null } | null, end?: { dateTime?: string | null, timeZone?: string | null } | null, attendees?: ({ displayName?: string | null, email?: string | null, responseStatus?: string | null })[] | null } }`\n\n#### listGoogleCalendarEvents\nList upcoming events from a Google Calendar, ordered by start time.\n- Requires a Google OAuth connection with Calendar events scope.\n- Only returns future events (timeMin = now).\n- The variable receives JSON or XML-like text depending on exportType. The direct execution output always returns structured events.\n- Input: `{ connectionId?: string, limit: number, exportType: "json" | "text", calendarId?: string }`\n- Output: `{ events: ({ id?: string | null, status?: string | null, htmlLink?: string | null, created?: string | null, updated?: string | null, summary?: string | null, description?: string | null, location?: string | null, organizer?: { displayName?: string | null, email?: string | null } | null, start?: { dateTime?: string | null, timeZone?: string | null } | null, end?: { dateTime?: string | null, timeZone?: string | null } | null, attendees?: ({ displayName?: string | null, email?: string | null, responseStatus?: string | null })[] | null })[] }`\n\n#### searchGoogleCalendarEvents\nSearch for events in a Google Calendar by keyword, date range, or both.\n- Requires a Google OAuth connection with Calendar events scope.\n- Supports keyword search via "query" and date filtering via "timeMin"/"timeMax" (ISO 8601 format).\n- Unlike "List Events" which only shows future events, this allows searching past events too.\n- Input: `{ query?: string, timeMin?: string, timeMax?: string, calendarId?: string, limit?: number, exportType: "json" | "text", connectionId?: string }`\n- Output: `{ events: ({ id?: string | null, status?: string | null, htmlLink?: string | null, created?: string | null, updated?: string | null, summary?: string | null, description?: string | null, location?: string | null, organizer?: { displayName?: string | null, email?: string | null } | null, start?: { dateTime?: string | null, timeZone?: string | null } | null, end?: { dateTime?: string | null, timeZone?: string | null } | null, attendees?: ({ displayName?: string | null, email?: string | null, responseStatus?: string | null })[] | null })[] }`\n\n#### updateGoogleCalendarEvent\nUpdate an existing event on a Google Calendar. Only specified fields are changed.\n- Requires a Google OAuth connection with Calendar events scope.\n- Fetches the existing event first, then applies only the provided updates. Omitted fields are left unchanged.\n- Attendees are specified as one email address per line, and replace the entire attendee list.\n- Input: `{ connectionId?: string, eventId: string, summary?: string, description?: string, location?: string, startDateTime?: string, endDateTime?: string, attendees?: string, calendarId?: string }`\n- Output: `{ eventId: string, htmlLink: string }`\n\n### Google Drive\n\n#### getGoogleDriveFile\nDownload a file from Google Drive and rehost it on the CDN. Returns a public CDN URL.\n- Requires a Google OAuth connection with Drive scope.\n- Google-native files (Docs, Sheets, Slides) cannot be downloaded \u2014 use dedicated steps instead.\n- Maximum file size: 200MB.\n- The file is downloaded and stored on the public CDN, where the returned URL is publicly accessible \u2014 pass `store` to put it in one of the app\'s own file stores instead, which for a private store means the URL only loads for an authorized session.\n- Input: `{ fileId: string, connectionId?: string }`\n- Output: `{ url: string, name: string, mimeType: string, size: number }`\n\n#### listGoogleDriveFiles\nList files in a Google Drive folder.\n- Requires a Google OAuth connection with Drive scope.\n- If folderId is omitted, lists files in the root folder.\n- Returns file metadata including name, type, size, and links.\n- Input: `{ folderId?: string, limit?: number, connectionId?: string, exportType: "json" | "text" }`\n- Output: `{ files: { id: string, name: string, mimeType: string, size: string, webViewLink: string, createdTime: string, modifiedTime: string }[] }`\n\n#### searchGoogleDrive\nSearch for files in Google Drive by keyword.\n- Requires a Google OAuth connection with Drive scope.\n- Searches file content and names using Google Drive\'s fullText search.\n- Input: `{ query: string, limit?: number, connectionId?: string, exportType: "json" | "text" }`\n- Output: `{ files: { id: string, name: string, mimeType: string, size: string, webViewLink: string, createdTime: string, modifiedTime: string }[] }`\n\n### HubSpot\n\n#### hubspotCreateCompany\nCreate a new company or update an existing one in HubSpot. Matches by domain.\n- Requires a HubSpot OAuth connection (connectionId).\n- If a company with the given domain already exists, it is updated. Otherwise, a new one is created.\n- Property values are type-checked against enabledProperties before being sent to HubSpot.\n- Input: `{ connectionId?: string, company: { domain: string, name: string }, enabledProperties: ({ label: string, value: string, type: "string" | "number" | "bool" })[] }`\n- Output: `{ companyId: string }`\n\n#### hubspotCreateContact\nCreate a new contact or update an existing one in HubSpot. Matches by email address.\n- Requires a HubSpot OAuth connection (connectionId).\n- If a contact with the given email already exists, it is updated. Otherwise, a new one is created.\n- If companyDomain is provided, the contact is associated with that company (creating the company if needed).\n- Property values are type-checked against enabledProperties before being sent to HubSpot.\n- Input: `{ connectionId?: string, contact: { email: string, firstname: string, lastname: string }, enabledProperties: ({ label: string, value: string, type: "string" | "number" | "bool" })[], companyDomain: string }`\n- Output: `{ contactId: string }`\n\n#### hubspotGetCompany\nLook up a HubSpot company by domain name or company ID.\n- Requires a HubSpot OAuth connection (connectionId).\n- Returns null if the company is not found.\n- When searching by domain, performs a search query then fetches the full company record.\n- Use additionalProperties to request specific HubSpot properties beyond the defaults.\n- Input: `{ connectionId?: string, searchBy: "domain" | "id", companyDomain: string, companyId: string, additionalProperties: string[] }`\n- Output: `{ company: { id: string, properties: object, createdAt: string, updatedAt: string, archived: boolean } | null }`\n\n#### hubspotGetContact\nLook up a HubSpot contact by email address or contact ID.\n- Requires a HubSpot OAuth connection (connectionId).\n- Returns null if the contact is not found.\n- Use additionalProperties to request specific HubSpot properties beyond the defaults.\n- Input: `{ connectionId?: string, searchBy: "email" | "id", contactEmail: string, contactId: string, additionalProperties: string[] }`\n- Output: `{ contact: { id: string, properties: object, createdAt: string, updatedAt: string, archived: boolean } | null }`\n\n### Hunter.io\n\n#### hunterApiCompanyEnrichment\nLook up company information by domain using Hunter.io.\n- Returns company name, description, location, industry, size, technologies, and more.\n- If the domain input is a full URL, the hostname is automatically extracted.\n- Returns null if the company is not found.\n- Input: `{ domain: string }`\n- Output: `{ data: { name: string, domain: string, description: string | null, country: string | null, state: string | null, city: string | null, industry: string | null, employees_range: string | null, logo_url: string | null, technologies: string[] } | null }`\n\n#### hunterApiDomainSearch\nSearch for email addresses associated with a domain using Hunter.io.\n- If the domain input is a full URL, the hostname is automatically extracted.\n- Returns a list of email addresses found for the domain along with organization info.\n- Input: `{ domain: string }`\n- Output: `{ data: { domain: string, disposable: boolean, webmail: boolean, accept_all: boolean, pattern: string, organization: string, country: string | null, state: string | null, emails: ({ value: string, type: string, confidence: number, first_name: string | null, last_name: string | null, position: string | null, seniority: string | null, department: string | null, linkedin: string | null, twitter: string | null, phone_number: string | null })[], linked_domains: string[] } }`\n\n#### hunterApiEmailFinder\nFind an email address for a specific person at a domain using Hunter.io.\n- Requires a first name, last name, and domain.\n- If the domain input is a full URL, the hostname is automatically extracted.\n- Returns the most likely email address with a confidence score.\n- Input: `{ domain: string, firstName: string, lastName: string }`\n- Output: `{ data: { first_name: string, last_name: string, email: string, score: number, domain: string, accept_all: boolean, position: string | null, twitter: string | null, linkedin_url: string | null, phone_number: string | null, company: string | null, sources: { domain: string, uri: string, extracted_on: string }[] } }`\n\n#### hunterApiEmailVerification\nVerify whether an email address is valid and deliverable using Hunter.io.\n- Checks email format, MX records, SMTP server, and mailbox deliverability.\n- Returns a status ("valid", "invalid", "accept_all", "webmail", "disposable", "unknown") and a score.\n- Input: `{ email: string }`\n- Output: `{ data: { status: string, result: string, score: number, email: string, regexp: boolean, gibberish: boolean, disposable: boolean, webmail: boolean, mx_records: boolean, smtp_server: boolean, smtp_check: boolean, accept_all: boolean, block: boolean, sources: { domain: string, uri: string, extracted_on: string }[] } }`\n\n#### hunterApiPersonEnrichment\nLook up professional information about a person by their email address using Hunter.io.\n- Returns name, job title, social profiles, and company information.\n- If the person is not found, returns an object with an error message instead of throwing.\n- Input: `{ email: string }`\n- Output: `{ data: { first_name: string, last_name: string, email: string, position: string | null, seniority: string | null, department: string | null, linkedin_url: string | null, twitter: string | null, phone_number: string | null, company: { name: string, domain: string, industry: string | null } | null } | { error: string } }`\n\n### LinkedIn\n\n#### postToLinkedIn\nCreate a post on LinkedIn from the connected account.\n- Requires a LinkedIn OAuth connection (connectionId).\n- Supports text posts, image posts, video posts, document posts, and article posts.\n- Attach one media type per post: image, video, document, or article.\n- Documents support PDF, PPT, PPTX, DOC, DOCX (max 100MB, 300 pages). Displays as a slideshow carousel.\n- Articles create a link preview with optional custom title, description, and thumbnail.\n- Visibility controls who can see the post.\n- Input: `{ message: string, visibility: "PUBLIC" | "CONNECTIONS", imageUrl?: string, videoUrl?: string, documentUrl?: string, articleUrl?: string, titleText?: string, descriptionText?: string, connectionId?: string }`\n- Output: `unknown`\n\n### Meshy\n\n#### meshyAnimate\nApply a preset animation to a rigged 3D character model using Meshy.\n- Requires a rig_task_id from a previously completed Meshy rigging step.\n- Select an animation from Meshy\'s library of 600+ preset animations.\n- Only works with humanoid (bipedal) rigged characters.\n- Supports post-processing: FPS change (24/25/30/60), FBX-to-USDZ conversion, or armature extraction.\n- Animation categories: DailyActions, WalkAndRun, Fighting, Dancing, BodyMovements.\n- Input: `{ rigTaskId: string, actionId: number }`\n- Output: `{ glbUrl: string, fbxUrl?: string, objUrl?: string, usdzUrl?: string, thumbnailUrl?: string, textureUrls?: object[], animations?: { name: string, glbUrl?: string, fbxUrl?: string }[], providerTaskId?: string }`\n\n#### meshyImageTo3d\nGenerate a 3D model from one or more images using Meshy. Uses the multi-image-to-3D endpoint.\n- Accepts 1-4 image URLs. All images should depict the same object from different angles for best results.\n- By default generates with textures. Set shouldTexture to false for mesh-only output.\n- Uses should_remesh: false to preserve UV mapping integrity.\n- Input: `{ imageUrls: string[], shouldTexture?: boolean, topology?: string, targetPolycount?: number, symmetryMode?: string, poseMode?: string, textureImageUrl?: string }`\n- Output: `{ glbUrl: string, fbxUrl?: string, objUrl?: string, usdzUrl?: string, thumbnailUrl?: string, textureUrls?: object[], animations?: { name: string, glbUrl?: string, fbxUrl?: string }[], providerTaskId?: string }`\n\n#### meshyRemesh\nRemesh an existing 3D model to adjust topology, polygon count, or convert formats using Meshy.\n- Provide either an input task ID (from a previous Meshy step) or a model URL.\n- Defaults to triangle topology with 30,000 target polys.\n- Useful for reducing face count before rigging (max 300k faces for rigging).\n- Input: `{ inputTaskId?: string, modelUrl?: string, topology?: string, targetPolycount?: number, resizeHeight?: number }`\n- Output: `{ glbUrl: string, fbxUrl?: string, objUrl?: string, usdzUrl?: string, thumbnailUrl?: string, textureUrls?: object[], animations?: { name: string, glbUrl?: string, fbxUrl?: string }[], providerTaskId?: string }`\n\n#### meshyRig\nAuto-rig a humanoid 3D model and generate basic walking/running animations using Meshy.\n- Only works well with standard humanoid (bipedal) models with clearly defined limbs.\n- Prefers model_url over input_task_id for cleaner rigging input.\n- Models with more than 300,000 faces should be remeshed first.\n- Returns rigged model files and optional basic animations.\n- Input: `{ inputTaskId?: string, modelUrl?: string, heightMeters?: number }`\n- Output: `{ glbUrl: string, fbxUrl?: string, objUrl?: string, usdzUrl?: string, thumbnailUrl?: string, textureUrls?: object[], animations?: { name: string, glbUrl?: string, fbxUrl?: string }[], providerTaskId?: string }`\n\n#### meshyTextTo3d\nGenerate a 3D model preview from a text prompt using Meshy. Produces an untextured mesh (preview stage).\n- Creates a text-to-3D preview task (mesh generation only, no texture).\n- Use the Meshy Texture step to apply textures to the preview.\n- Maximum prompt length is 600 characters.\n- Input: `{ prompt: string, modelType?: string, topology?: string, targetPolycount?: number, symmetryMode?: string, poseMode?: string }`\n- Output: `{ glbUrl: string, fbxUrl?: string, objUrl?: string, usdzUrl?: string, thumbnailUrl?: string, textureUrls?: object[], animations?: { name: string, glbUrl?: string, fbxUrl?: string }[], providerTaskId?: string }`\n\n#### meshyTexture\nApply or replace textures on a 3D model using a text prompt or reference image via Meshy.\n- Provide either an input task ID (from a previous Meshy step) or a model URL.\n- Provide either a text style prompt or an image style URL to guide texturing.\n- Supports .glb, .gltf, .obj, .fbx, .stl model formats when using modelUrl.\n- By default preserves original UVs (enableOriginalUv = true).\n- Works with any model source: text-to-3D previews, image-to-3D, remeshed models, or external files.\n- Input: `{ inputTaskId?: string, modelUrl?: string, textStylePrompt?: string, imageStyleUrl?: string, enableOriginalUv?: boolean, enablePbr?: boolean }`\n- Output: `{ glbUrl: string, fbxUrl?: string, objUrl?: string, usdzUrl?: string, thumbnailUrl?: string, textureUrls?: object[], animations?: { name: string, glbUrl?: string, fbxUrl?: string }[], providerTaskId?: string }`\n\n### Notion\n\n#### notionCreatePage\nCreate a new page in Notion as a child of an existing page.\n- Requires a Notion OAuth connection (connectionId).\n- Content is provided as markdown and converted to Notion blocks (headings, paragraphs, lists, code, quotes).\n- The page is created as a child of the specified parent page (pageId).\n- Input: `{ pageId: string, content: string, title: string, connectionId?: string }`\n- Output: `{ pageId: string, pageUrl: string }`\n\n#### notionUpdatePage\nUpdate the content of an existing Notion page.\n- Requires a Notion OAuth connection (connectionId).\n- Content is provided as markdown and converted to Notion blocks.\n- "append" mode adds content to the end of the page. "overwrite" mode deletes all existing blocks first.\n- Input: `{ pageId: string, content: string, mode: "append" | "overwrite", connectionId?: string }`\n- Output: `{ pageId: string, pageUrl: string }`\n\n### QuickBooks\n\n#### quickbooksApiRequest\nCall any QuickBooks Online Accounting API route using a connected company.\n- Requires a QuickBooks OAuth connection (connectionId).\n- Route is relative to /v3/company/{realmId}/, e.g. "account", "account/93", "query", "invoice".\n- Use queryParams for query-string values such as query=select * from Account.\n- Body is JSON for POST, PUT, and PATCH.\n- Input: `{ connectionId?: string, method: string, route: string, queryParams: object, body: string }`\n- Output: `{ ok: boolean, status: number, body: unknown }`\n\n### TypeSafe\n\n#### typesafeEvaluate\nAsk TypeSafe structured questions about text or JSON and get calibrated answers. The model returns a probability, a selected option, or a score. It does not write a reply. Each question is a noul (yes or no, as a probability from 0 to 1), a choice (one option from a set you define, with a probability for each), or a score (a rating against ordered levels). Put every question in the same call. The question id is yours: answers come back under the same ids. The model defaults to Jev (jev-latest).\n- state is the content to judge: a string, or a JSON object or array (a record, a chat log, application state). Text only. Turn images, audio, and video into text first.\n- questions is a map. The key is an id you choose. It is not shown to the model. The answer comes back under the same id.\n- A noul asks a yes/no question and returns noul, a probability from 0 (no) to 1 (yes). Optional criteria.true and criteria.false say what a yes and a no mean.\n- A choice picks one option. criteria is a map of option id to a short description (or null). At most 255 options. The answer includes the winning choice, a probability for every option, and confidence.\n- A score rates the state on an ordered rubric. criteria is 2 to 10 level descriptions, low to high. The answer\'s score can land between levels, and includes a probability per level plus confidence.\n- instructions may be a string, or an object that keeps the question in one field and reference data in the others. Refer to those fields by name in backticks.\n- Send every question you need in one call. The model reads the state once and answers them together.\n- model defaults to jev-latest, the current stable release. jev-preview tracks the newest release. Pin a version such as jev-1.13.0 when a threshold was tuned against that version. The response model field is the version that answered.\n- Billed per input token. Output tokens are free.\n- Input: `{ state: string | object | unknown[], questions: object, model?: string }`\n- Output: `{ model: string, answers: object, usage: { input_tokens: number, output_tokens: number } }`\n\n### X\n\n#### postToX\nCreate a post on X (Twitter) from the connected account.\n- Requires an X OAuth connection (connectionId).\n- Maximum 280 characters of text.\n- Optionally attach up to 4 media items (images, GIFs, or videos) via mediaUrls.\n- Media URLs must be publicly accessible. The service fetches and uploads them to X.\n- Supported formats: JPEG, PNG, GIF, WEBP, MP4. Images up to 5MB, videos up to 512MB.\n- Input: `{ text: string, connectionId?: string, mediaUrls?: string[] }`\n- Output: `unknown`\n\n#### searchXPosts\nSearch recent X (Twitter) posts matching a query.\n- Searches only the past 7 days of posts.\n- Query supports X API v2 search operators (up to 512 characters).\nAvailable search operators in query:\n| Operator         | Description                                      |\n| -----------------| -------------------------------------------------|\n| from:            | Posts from a specific user (e.g., from:elonmusk) |\n| to:              | Posts sent to a specific user (e.g., to:NASA)    |\n| @                | Mentions a user (e.g., @openai)                  |\n| #                | Hashtag search (e.g., #AI)                       |\n| is:retweet       | Filters retweets                                 |\n| is:reply         | Filters replies                                  |\n| has:media        | Posts containing media (images, videos, or GIFs) |\n| has:links        | Posts containing URLs                            |\n| lang:            | Filters by language (e.g., lang:en)              |\n| -                | Excludes specific terms (e.g., -spam)            |\n| ()               | Groups terms or operators (e.g., (AI OR ML))     |\n| AND, OR, NOT     | Boolean logic for combining or excluding terms   |\nConjunction-Required Operators (must be combined with a standalone operator):\n| Operator     | Description                                    |\n| ------------ | -----------------------------------------------|\n| has:media  | Posts containing media (images, videos, or GIFs) |\n| has:links  | Posts containing URLs                            |\n| is:retweet | Filters retweets                                 |\n| is:reply   | Filters replies                                  |\nFor example, has:media alone is invalid, but #AI has:media is valid.\n- Input: `{ query: string, scope: "recent" | "all", options: { startTime?: string, endTime?: string, maxResults?: number } }`\n- Output: `{ posts: { id: string, authorId: string, dateCreated: string, text: string, stats: { retweets: number, replies: number, likes: number } }[] }`\n\n### You.com\n\n#### youDotComFinanceResearch\nAsk a financial research question using You.com Finance Research and return the sourced response.\n- Use this for financial questions such as company analysis, earnings, market research, filings, macroeconomics, and due diligence.\n- researchEffort supports deep (default) or exhaustive.\n- Finance Research returns the same response shape as Web Research, but searches a finance-optimized index.\n- Use it for cited synthesis, not raw price feeds or structured time-series exports.\n- Input: `{ input: string, researchEffort?: "deep" | "exhaustive" }`\n- Output: `{ data: object | unknown[] }`\n\n#### youDotComGetPageContent\nFetch clean Markdown, HTML, or metadata for known URLs using the You.com Contents API.\n- Use this step when you already know the URLs. Use Web Search with livecrawl when You.com should discover pages from a query.\n- A single request supports up to 10 URLs.\n- Request only the formats you need. Markdown is recommended for LLM consumption.\n- Increase crawlTimeout for JavaScript-heavy pages, up to 60 seconds.\n- Individual pages can partially fail; check each returned item before processing.\n- Input: `{ urls: string[], formats?: ("markdown" | "html" | "metadata")[], crawlTimeout?: number }`\n- Output: `{ data: object | unknown[] }`\n\n#### youDotComLiveNews\nFetch live news articles through the You.com Search API and return the full structured response.\n- Defaults freshness to day for breaking or recent news.\n- Use country and language together to monitor regional or non-English news.\n- Use livecrawl: \'news\' with livecrawlFormats: [\'markdown\'] when you need full article text.\n- Use a custom freshness range like YYYY-MM-DDtoYYYY-MM-DD for historical news windows.\n- Input: `{ query: string, freshness?: string, count?: number, country?: string, language?: string, safesearch?: "off" | "moderate" | "strict", livecrawl?: "news" | "all", livecrawlFormats?: ("markdown" | "html")[] }`\n- Output: `{ data: object | unknown[] }`\n\n#### youDotComWebResearch\nAsk a research question and return a grounded You.com Research API answer with sources.\n- Use Web Search when you need raw URLs and snippets. Use Web Research when you want a synthesized answer with citations.\n- researchEffort controls depth and latency: lite, standard, deep, or exhaustive. standard is a good default.\n- sourceControl can restrict, exclude, or boost domains, and can apply freshness or country filters.\n- includeDomains cannot be combined with excludeDomains or boostDomains.\n- outputSchema returns structured output.content and is supported by standard, deep, and exhaustive, not lite.\n- Input: `{ input: string, researchEffort?: "lite" | "standard" | "deep" | "exhaustive", sourceControl?: { includeDomains?: string[], excludeDomains?: string[], boostDomains?: string[], freshness?: string, country?: string } | string, outputSchema?: object | string }`\n- Output: `{ data: object | unknown[] }`\n\n#### youDotComWebSearch\nSearch the web and news using the You.com Search API and return the full structured response.\n- Query supports You.com search operators:\n| Operator | Description | Example |\n| -------- | ----------- | ------- |\n| site: | Search within a domain and its subdomains | site:uscourts.gov |\n| filetype: | Search for a specific file type | filetype:pdf |\n| + | Require the exact term after the operator | +GAAP |\n| - | Exclude the exact term after the operator | -prs |\n| AND | Require both expressions | guitar AND Fender |\n| OR | Match either expression | guitar OR drum |\n| NOT | Negate an expression | NOT site:uscourts.gov |\n- Use livecrawl with livecrawlFormats: [\'markdown\'] when you need full page content instead of snippets.\n- Use the Get Page Content step when you already know the URLs to fetch.\n- Use freshness for recency: day, week, month, year, or YYYY-MM-DDtoYYYY-MM-DD.\n- Use country (ISO 3166-1 alpha-2) and language (BCP 47) to focus results geographically or linguistically.\n- includeDomains cannot be combined with excludeDomains or boostDomains.\n- Input: `{ query: string, count?: number, freshness?: string, country?: string, language?: string, offset?: number, safesearch?: "off" | "moderate" | "strict", livecrawl?: "web" | "news" | "all", livecrawlFormats?: ("markdown" | "html")[], crawlTimeout?: number, includeDomains?: string[], excludeDomains?: string[], boostDomains?: string[] }`\n- Output: `{ data: object | unknown[] }`\n\n### YouTube\n\n#### fetchYoutubeCaptions\nRetrieve the captions/transcript for a YouTube video.\n- Supports multiple languages via the language parameter.\n- "text" export produces timestamped plain text; "json" export produces structured transcript data.\n- Input: `{ videoUrl: string, exportType: "text" | "json", language: string }`\n- Output: `{ transcripts: { text: string, start: number }[] }`\n\n#### fetchYoutubeChannel\nRetrieve metadata and recent videos for a YouTube channel.\n- Accepts a YouTube channel URL (e.g. https://www.youtube.com/@ChannelName or /channel/ID).\n- Returns channel info and video listings as a JSON object.\n- Input: `{ channelUrl: string }`\n- Output: `object`\n\n#### fetchYoutubeComments\nRetrieve comments for a YouTube video.\n- Paginates through comments (up to 5 pages).\n- "text" export produces markdown-formatted text; "json" export produces structured comment data.\n- Input: `{ videoUrl: string, exportType: "text" | "json", limitPages: string }`\n- Output: `{ comments: { id: string, link: string, publishedDate: string, text: string, likes: number, replies: number, author: string, authorLink: string, authorImg: string }[] }`\n\n#### fetchYoutubeVideo\nRetrieve metadata for a YouTube video (title, description, stats, channel info).\n- Returns video metadata, channel info, and engagement stats.\n- Video format data is excluded from the response.\n- Input: `{ videoUrl: string }`\n- Output: `object`\n\n#### searchYoutube\nSearch for YouTube videos by keyword.\n- Supports pagination (up to 5 pages) and country/language filters.\n- Use the filter/filterType fields for YouTube search parameter (sp) filters.\n- Input: `{ query: string, limitPages: string, filter: string, filterType: string, countryCode?: string, languageCode?: string }`\n- Output: `{ results: object }`\n\n#### searchYoutubeTrends\nRetrieve trending videos on YouTube by category and region.\n- Categories: "now" (trending now), "music", "gaming", "films".\n- Supports country and language filtering.\n- Input: `{ bp: "now" | "music" | "gaming" | "films", hl: string, gl: string }`\n- Output: `object`\n\n### Helpers\n\n#### `listModels()`\nList all available AI models across all categories.\n\nOutput:\n```typescript\n{\n  models: {\n    id: string;\n    name: string;            // Display name\n    type: "llm_chat" | "image_generation" | "video_generation" | "video_analysis" | "text_to_speech" | "vision" | "transcription";\n    maxTemperature: number;\n    maxResponseSize: number;\n    inputs: object[];        // Accepted input types\n  }[]\n}\n```\n\n#### `listModelsByType(modelType)`\nList AI models filtered by type.\n- `modelType`: `"llm_chat"` | `"image_generation"` | `"video_generation"` | `"video_analysis"` | `"text_to_speech"` | `"vision"` | `"transcription"`\n- Output: same as `listModels()`\n\n#### `listModelsSummary()`\nList all available AI models (summary). Returns only id, name, type, and tags. Suitable for display or consumption inside a model context window.\n\nOutput:\n```typescript\n{\n  models: {\n    id: string;\n    name: string;\n    type: "llm_chat" | "image_generation" | "video_generation" | "video_analysis" | "text_to_speech" | "vision" | "transcription";\n    tags: string;            // Comma-separated tags\n  }[]\n}\n```\n\n#### `listModelsSummaryByType(modelType)`\nList AI models (summary) filtered by type.\n- `modelType`: `"llm_chat"` | `"image_generation"` | `"video_generation"` | `"video_analysis"` | `"text_to_speech"` | `"vision"` | `"transcription"`\n- Output: same as `listModelsSummary()`\n\n#### `listConnectors()`\nList available OAuth connector services (Slack, Google, HubSpot, etc.) and their actions. These are third-party integrations \u2014 for most tasks, use actions directly instead.\n\nOutput:\n```typescript\n{\n  services: {\n    id: string;\n    name: string;\n    icon: string;\n    actions: { id: string; name: string }[];\n  }[]\n}\n```\n\n#### `getConnector(serviceId)`\nGet details for a single OAuth connector service by ID.\n\nOutput:\n```typescript\n{\n  service: {\n    id: string;\n    name: string;\n    icon: string;\n    actions: { id: string; name: string }[];\n  }\n}\n```\n\n#### `getConnectorAction(serviceId, actionId)`\nGet the full configuration for an OAuth connector action, including all input fields needed to call it via `runFromConnectorRegistry`. OAuth connectors are sourced from the open-source MindStudio Connector Registry (MSCR) with 850+ actions across third-party services.\n\nOutput:\n```typescript\n{\n  action: {\n    id: string;\n    name: string;\n    description: string;\n    quickHelp: string;\n    configuration: { title: string; items: { label: string; helpText: string; variable: string; type: string; defaultValue: string; placeholder: string; selectOptions?: object }[] }[];\n  }\n}\n```\n\n#### `listConnections()`\nList OAuth connections for the organization (authenticated third-party service links). Use the returned connection IDs when calling OAuth connector actions. Connectors require the user to connect to the third-party service in MindStudio before they can be used.\n\nOutput:\n```typescript\n{\n  connections: {\n    id: string;       // Connection ID to pass to connector actions\n    provider: string; // Integration provider (e.g. slack, google)\n    name: string;     // Display name or account identifier\n  }[]\n}\n```\n\n#### `estimateStepCost(stepType, step?, options?)`\nEstimate the cost of executing a step before running it. Pass the same step config you would use for execution.\n\n```typescript\nconst estimate = await mindstudio.estimateStepCost(\'generateText\', { message: \'Hello\' });\n```\n\n- `stepType`: string \u2014 The action name (e.g. `"generateText"`).\n- `step`: object \u2014 Optional action input parameters for more accurate estimates.\n- `options`: `{ appId?: string, workflowId?: string }` \u2014 Optional context for pricing.\n\nOutput:\n```typescript\n{\n  costType?: string;  // "free" when the step has no cost\n  estimates?: {\n    eventType: string;       // Billing event type\n    label: string;           // Human-readable cost label\n    unitPrice: number;       // Price per unit in nanodollars (1e-9 USD)\n    unitType: string;        // What constitutes a unit (e.g. "token", "request")\n    estimatedCost?: number;  // Estimated total cost in nanodollars, or null if not estimable\n    quantity: number;        // Number of billable units\n  }[]\n}\n```\n\n#### `changeName(displayName)`\nUpdate the display name of the authenticated agent. Useful for agents to set their own name after connecting.\n\n```typescript\nawait mindstudio.changeName(\'My Agent\');\n```\n\n#### `changeProfilePicture(profilePictureUrl)`\nUpdate the profile picture of the authenticated agent. Useful for agents to set their own avatar after connecting.\n\n```typescript\nawait mindstudio.changeProfilePicture(\'https://example.com/avatar.png\');\n```\n\n#### `uploadFile(content, options)`\n**Deprecated for app file storage** \u2014 use the `files` store below (`files.defineStore(...).put(...)`): private by default, app-scoped, and served on the app\'s own domain. `uploadFile` uploads to the shared account media CDN and remains only for account-level assets (e.g. an agent avatar for `changeProfilePicture`). Gets a signed upload URL, PUTs the file content, and returns the permanent public URL.\n\n```typescript\nimport { readFileSync } from \'fs\';\nconst { url } = await mindstudio.uploadFile(readFileSync(\'photo.png\'), { extension: \'png\', type: \'image/png\' });\n```\n\n- `content`: `Buffer | Uint8Array` \u2014 The file content.\n- `options.extension`: string \u2014 File extension without the dot (e.g. `"png"`, `"jpg"`, `"mp4"`).\n- `options.type`: string (optional) \u2014 MIME type (e.g. `"image/png"`). Determines which CDN subdomain is used.\n\nOutput: `{ url: string }` \u2014 The permanent public CDN URL.\n\n#### `files` \u2014 per-app file storage (private by default)\nTyped per-app file storage: user uploads, generated documents, images, marketing assets. Think of a store as a CDN-backed bucket the app talks to \u2014 NOT app-defined state like a `db` table. The API is *shaped* like the `db` namespace (define a store at module scope, then read/write), but its contents aren\'t modeled by code, and one store is shared across dev and prod (a file uploaded in the dev editor is already live in prod at the same URL). Creates are safe by default because keys default to unique (UUID, or a content-addressed hash); the operations to treat with care are `delete(key)` and overwriting a fixed key. Public objects are CDN-cached, and the CDN drops one when it is overwritten or deleted; its `Cache-Control`, set at put time via `cacheControl`, is how long a browser may keep a copy, at most an hour: auto-minted keys get the hour (the key is never reused), and named keys default to `public, max-age=300`, so an overwrite reaches every browser within ~5 minutes; pass `cacheControl` to tune (e.g. `\'public, max-age=60\'` for near-live config, `\'no-store\'` to revalidate every read). Files are served on the app\'s own domain: private stores (the default) are signed / session-authorized; public stores are world-readable, CDN-served, and images resize via query params (e.g. `?w=400&h=300&fit=cover`). Prefer this over the deprecated `uploadFile()` for app storage.\n\n```typescript\nimport { files } from \'@mindstudio-ai/agent\';\n\n// Define stores at module scope (access is pinned here).\nexport const Uploads = files.defineStore(\'uploads\');                  // private (default)\nexport const Assets  = files.defineStore(\'assets\', { access: \'public\' });\n\n// Store bytes and get a ready-to-use URL.\nconst file = await Uploads.put(buffer, { contentType: \'application/pdf\', filename: \'report.pdf\' });\nfile.url;              // stable on-domain URL for a logged-in user (drop into <img>/<a download>)\nawait file.shareUrl(); // absolute signed URL that works with NO session (email/embed); default 24h\n```\n\nStore methods:\n- `put(content, { key?, contentType?, filename?, contentAddressed?, cacheControl? })` \u2192 `StoredFile`. `content` is `Buffer | Uint8Array | string` \u2014 any size up to 5 GiB (large content transparently uploads direct to storage via a presigned POST, so there is no request-body cap). Omit `key` for a random UUID; `contentAddressed: true` derives an immutable `<sha256>.<ext>` key. `cacheControl` sets how long a browser may keep a public object, at most an hour (defaults: auto-minted keys an hour, named keys `public, max-age=300`).\n- `get(key)` \u2192 `Buffer` \xB7 `head(key)` \u2192 `StoredFile` \xB7 `exists(key)` \u2192 `boolean`\n- `list({ prefix?, cursor?, limit? })` \u2192 `{ files: StoredFile[]; cursor? }`\n- `delete(key)` \u2192 `void`\n- `shareUrl(key, { expiresIn? })` \u2192 signed absolute URL (private stores; no session needed)\n\n`StoredFile`: `{ store, key, access, size?, contentType?, updatedAt?, url, shareUrl() }`.\n\n`defineStore(name, options?)` options: `access: \'public\' | \'private\'` (default private), `maxSize` (bytes; cap for client-direct uploads), `contentTypes` (allowlist for client-direct uploads).\n\nUser uploads (client-direct \u2014 bytes go straight to storage, never through the app backend): a backend method mints a token, the frontend submits it with `@mindstudio-ai/interface`.\n```typescript\n// backend method\nconst token = await Uploads.createUploadToken({ contentType, maxSize: 25 * 1024 * 1024 }); // also accepts cacheControl\nreturn token;\n// frontend (@mindstudio-ai/interface)\nimport { platform } from \'@mindstudio-ai/interface\';\nconst { url } = await platform.upload(token, file, { onProgress });\n```\n\nGenerated assets: EVERY action that produces a file can put its output straight into one of these stores \u2014 pass the store handle as `store` in the OPTIONS object (second argument). There is no allowlist; if an action returns a file URL, it takes `store`. Optional: omit it and the asset goes to the shared MindStudio CDN as before. Reach for it when the asset belongs to the app \u2014 private to a user, or listed alongside its other files. Inputs are never published either way: a file you hand an action is read where it sits, and any intermediate the platform needs is written to private storage and dropped when the call finishes.\n\n```typescript\nconst { imageUrl } = await mindstudio.generateImage({ prompt: \'a red sneaker\' }, { store: Assets });\n```\n\n#### `db` \u2014 typed tables over the app\'s managed SQLite\n\nChainable, typed collection API over the app\'s managed databases. Define a table at module scope; every read/write is lazy until awaited, and reads/writes can be batched into one round trip with `db.batch()`.\n```typescript\nimport { db } from \'@mindstudio-ai/agent\';\n\nexport const Orders = db.defineTable<Order>(\'orders\');\n\n// Reads \u2014 filters compile to SQL WHERE when possible\nconst recent = await Orders.filter(o => o.status === \'paid\').sortBy(o => o.created_at).reverse().take(20);\n// Comparing to an outer-scope value? Use the bindings form so it compiles to SQL:\nconst mine = await Orders.filter((o, $) => o.requestedBy === $.uid, { uid: auth.userId }); // bindings: lifts closure var so filter compiles to SQL\n\n// Writes\nconst order = await Orders.push({ item: \'Laptop\', amount: 999, status: \'pending\' });\nawait Orders.update(order.id, { status: \'approved\' });\n```\n\nAggregation \u2014 `count()`, `sum(fn)`, `avg(fn)`, `countDistinct(fn)`, and `aggregate(spec)` compile to SQL aggregates (COUNT/TOTAL/AVG/GROUP BY); no rows are fetched, so they stay cheap at any table size. Never page a whole table into memory to compute a summary.\n```typescript\nconst top = await Answers\n  .filter((a, $) => a.surveyId === $.surveyId, { surveyId }) // bindings: compiles to SQL\n  .aggregate({\n    by: [\'questionId\', \'dimension\'],\n    select: { n: { count: true }, total: { sum: \'score\' }, avgScore: { avg: \'score\' }, respondents: { countDistinct: \'responseId\' } },\n    orderBy: \'total\', desc: true, limit: 20,\n  });\n// Array<{ questionId, dimension, n: number, total: number, avgScore: number | null, respondents: number }>\n```\nSelect terms: `{ count: true }` \xB7 `{ sum: \'col\' }` \xB7 `{ avg: \'col\' }` \xB7 `{ min: \'col\' }` \xB7 `{ max: \'col\' }` \xB7 `{ countDistinct: \'col\' }` (string column names). Omit `by` for whole-set aggregates (single object). Empty set: count/countDistinct 0, sum 0, avg/min/max null.\n\nRaw SQL escape hatch \u2014 `db.sql<T>(query, params?, { database? })` runs read-only raw SQL (joins, subqueries, window functions) against the app\'s own managed database. SELECT/WITH only; writes throw \u2014 use Table methods. Lazy and batchable. Last resort: prefer the typed API (including `aggregate()`) whenever it can express the query; raw rows may not exactly match the typed API\'s value representations.\n```typescript\nconst rows = await db.sql<{ questionId: string; n: number }>(\'SELECT questionId, COUNT(*) AS n FROM answers WHERE surveyId = ? GROUP BY questionId\', [surveyId]);\n```\n\n#### `dataSources` \u2014 searchable document corpora (RAG)\n\nDeclare a data source at module scope and import the handle. The platform owns parsing, chunking, embedding, storage and tenant isolation; every hit returns a citation.\n```typescript\nimport { dataSources } from \'@mindstudio-ai/agent\';\nexport const Policies = dataSources.defineDataSource(\'policies\');\n\n// search\nconst { results } = await Policies.search(\'what are the payment terms?\', { topK: 5 });\n// options: topK (default 5, max 50), scoreThreshold, filter, mode, maxPerDocument, highlight, rerank, hybrid\nconst context = results.map((r) => r.text).join(String.fromCharCode(10, 10));\n\n// narrow before ranking: metadata tags, filename, pages, required words\nawait Policies.search(\'termination clause\', {\n  filter: { metadata: { department: \'legal\', signedAt: { gte: 20250101, lte: 20260630 } }, phrase: \'notice period\' },\n  maxPerDocument: 2,  // stop one document monopolizing the results\n});\n\n// mode: \'hybrid\' (default) | \'semantic\' | \'lexical\'. Lexical skips the query\n// embedding \u2014 fastest, right for identifier-shaped queries (error codes, SKUs).\nawait Policies.search(\'ERR-7741X\', { mode: \'lexical\' });\n\n// add a document (background \u2014 poll documents() for status)\nawait Policies.add(buffer, { filename: \'policy.pdf\', contentType: \'application/pdf\', metadata: { department: \'legal\' } });\nconst docs = await Policies.documents();\n\n// walk a large corpus (a page at a time behind the scenes) to build your own index of it\nfor await (const doc of Policies.allDocuments()) { /* doc.id, doc.metadata, doc.status */ }\n\n// introspection \u2014 what is in the corpus, and how one document was split\nconst { chunkCount, pipeline } = await Policies.stats();  // counts + the config actually in effect\nconst chunks = await Policies.chunks(docs[0].id);       // why a document does or doesn\'t match\n\n// how many passages contain the words \u2014 exact, same filter grammar as search. Say\n// "N passages mention these words", never "N relevant results": no such total exists,\n// and search\'s hits are not this set (semantic hits need not contain the words).\nconst { chunks: mentions } = await Policies.count({ contains: \'client secret rotation\' });\n\n// what the matching passages are about: your own ids, asserted by add(), a mapper or a manifest\nconst { entities } = await Policies.entities({ filter: { entities: [\'vendor-acme\'] } });  // [{ id, chunks }], most first\n\n// the source\'s dictionary: your ids and their names, so passages are tagged wherever the text names them\nawait Policies.pushDictionary(vendors.map((v) => ({ id: v.id, names: [v.name, ...v.aliases] })));\nconst { mentions } = await Policies.link(\'Renewal terms agreed with ACME Corp.\');  // what it finds: [{ ids, text }]\n```\n\n`SearchHit`: `{ score, text, entities, citation: { documentId, filename, pageNumber, chunkIndex, headingPath, boundingBox?, url } }`. `citation.url` is a stable on-domain link to the source document for a signed-in user \u2014 drop it in an `<a href>`. A document is a cousin of a file and shares its share verbs: for a public app or another origin, `search(q, { shareCitations: 3600 })` returns every `citation.url` as an absolute signed link that needs no session, and `Policies.shareUrl(hit.citation, { expiresIn })` signs one document on demand (default 24h). `(documentId, chunkIndex)` is a chunk\'s stable identity \u2014 key an eval set on that, not on `text`.\n\nAlso on each hit: `retrievalRank`/`retrievalScore` (position before reranking \u2014 compare with the final position to see what reranking did). Three opt-in search options for debugging and display: `explain: true` adds `explain.{dense,lexical,matchedVia}` so you can see which half of hybrid found a hit; `expand: 1` adds `neighbors.{before,after}` for surrounding context; `highlight: true` adds `matches` \u2014 `{start, end}` offsets of query terms in `text`, for rendering highlights. None of them changes the results or their order.\n\nScore semantics: `score`\'s scale depends on the mode \u2014 cosine similarity (`semantic`), a rank-fusion reciprocal (`hybrid`; small numbers, not similarities), keyword-overlap weight (`lexical`) \u2014 and when reranking ran, it\'s the reranker\'s 0\u20131 relevance instead (the retrieval value stays in `retrievalScore`). So with reranking on (the default), put quality cutoffs on `score`. `scoreThreshold` cuts on the retrieval branch BEFORE fusion/reranking \u2014 a per-embedding-model cosine floor, not a relevance cutoff \u2014 so leave it unset unless you\'ve measured it on your corpus.\n\n`filter` fields (all AND together, and only narrow): `metadata` (per-key equals, any-of via an array, or a numeric range via `{gte?, lte?}`), `filename`, `documentIds`, `entities` (any of these ids), `pages: {min?, max?}`, `contains` (all words, any order), `phrase` (exact adjacent sequence). Document metadata is set at add time: scalars only, \u226416 keys, keys `[a-zA-Z0-9_-]`. Ranges are numeric only \u2014 to filter by date, store dates as sortable integers (`20250315`-style YYYYMMDD, or epoch seconds) and range on those. Re-adding the same bytes with different metadata or entities updates them in place \u2014 no re-processing.\n\nA document\'s `entities` name what it is about as the app\'s own ids (an asset tag, a person\'s row id), whether or not its text says so: set by `add(bytes, { entities })`, a mapper\'s `entities` or a job manifest line\'s, up to 256 per document, each 1\u2013128 printable characters with no spaces or commas, and carried by every passage. Metadata describes a document; entities say what it is about, and the app\'s own tables say how those things relate. `filter.entities` narrows to passages about any of the ids. `entities({ filter?, limit? })` lists which entities the matching passages are about, most first, as `{ id, chunks }`: approximate counts of passages, not documents (limit default 20, max 1,000; `filter: { documentIds: [id] }` for one document\'s). Filter on one entity to see what appears beside it.\n\nA source\'s DICTIONARY tags passages by what their text says: `pushDictionary(entries)` sends every `{ id, names }` the app has (a projection of its own tables: up to 250,000 entries, 16 names each, a million names in all) and replaces the dictionary; from then on every passage carries the ids whose names it mentions, beside its document\'s asserted ones, and `filter.entities`, hits and `entities()` see both. Matching is deterministic and model-free: case and a tag\'s separators don\'t matter (`P-101A`, `P101A`, `p 101a`), nor a size\'s inch or foot mark before one (`8"-P-1057-A1A` is `8-P-1057-A1A`), a name matches only as whole words (`P-101AB` is not `P-101A`), the longest name wins, and a name two ids claim tags both. A name needs three letters or digits (two in Chinese, Japanese or Korean). Passages indexed before a push are re-tagged in the background (a relink, nothing re-embedded; `dictionary()` shows it). Push it whole after each load of the tables it comes from: an unchanged dictionary commits nothing (`unchanged: true`). The result reports `ambiguousNames`, `mostClaimed` (each `{ key, name, ids }`: the folded key and one name as pushed) and names that can\'t match their own text (`unmatchable`). `link(text)` shows what the current dictionary finds in any text; use it to check a dictionary before relying on it. A mapper may not push one.\n\nSearch is deterministic for a fixed corpus and configuration \u2014 same query, same order \u2014 so regression checks are meaningful. `search()` also returns `latencyMs`.\n\n`search()` throws `index_warming` (HTTP 503) when a large corpus\'s index is being reloaded after an eviction \u2014 a minute or two, in the background. It means loading, not empty: tell the user the knowledge base is warming up and retry shortly. Small corpora reload inside the search and never raise it; sources on dedicated capacity are never evicted.\n\nOn dedicated capacity `search()` throws `index_building` (HTTP 503) instead, while a bulk load runs and while the index is built after one: the index is built once at the end of a load rather than behind every write, and the message carries "N of M vectors indexed". Handle it like `index_warming` \u2014 the knowledge base is being built, never empty; say so and retry later.\n\nOther methods: `stats()`, `documents()` (the first thousand, or `documents({ ids })` to poll what `add` returned), `documentsPage({ cursor?, limit?, order? })` (one page; `order: "newest"` is what just arrived) and `allDocuments()` (walk a corpus of any size, oldest first; to write what each document says into the app\'s own tables, give the source a READER instead), `passages({ filter?, cursor?, limit? })` and `allPassages(filter?)` (every passage matching a filter, unranked, document by document, as `{ text, entities, citation }`: every page naming an entity rather than search\'s best few; limit default 50, max 200), `chunks(documentId, {vectors?})`, `remove(documentId)`, `removeWhere(selector)`, `ensure(name?)` (rarely needed \u2014 `add` and `search` create-on-reference), and `DataSource.contentHash(bytes)` to check for an existing document before adding.\n\n`removeWhere({ metadata?, filename?, documentIds?, externalIdPrefix? })` removes every matching document (vectors, row and bytes) in pages of a thousand and returns `{ deleted, remaining }`. Chunk-level filter fields, `entities` and an empty selector are refused; deleting a whole source is an owner operation (`remy-admin datasources delete`), not something app code does.\n\nKeeping a corpus in sync with an S3 bucket: the owner connects the bucket once with `remy-admin datasources connect --source <slug> --bucket <b> --region <r> --prefix <p> --access-key-secret <NAME> --secret-key-secret <NAME>` (the two names are app secrets set with `remy-admin secrets set`; credentials are never written in code or pasted in chat). From then on `await Source.sync()` lists the bucket, diffs it by ETag against what was ingested, and runs a job over what changed \u2014 returning `{ job }` immediately; poll `Source.job(id)` or list `Source.jobs()`. A nightly sync is an ordinary cron interface job whose method calls `sync()`; nothing is spent when nothing changed. The job auto-approves under the connector\'s per-sync budget; a plan over it waits in `planned` for the owner to `remy-admin datasources jobs approve <id>`.\n\nWhen the file is not the document \u2014 a JSON record that should become markdown plus metadata, a JSONL object holding many articles, a kill notice that means "remove this story" \u2014 give the source a MAPPER: `datasources/<slug>.mapper.ts` exporting `defineMapper(Source, { map })` from `@mindstudio-ai/agent`, declared in mindstudio.json as `"dataSources": [{ "slug": "<slug>", "mapper": { "path": "dist/datasources/<slug>.mapper.ts" } }]`. `map(object)` gets `{ key, size, contentType, etag, lastModified, metadata, entities, bytes(), text(), json() }` and returns one of `documents([{ externalId, title, markdown, metadata?, entities?, replaces? }])`, `passthrough({ metadata?, entities? })` (ingest the raw object as-is), `skip(reason)`, or `deletes([externalId, ...])`; throwing is the object\'s error. `documents()` takes an array or any iterable or async iterable with no cap on its length, so a generator that yields a document per row streams an export of any size; a document that fails validation is left out and named, the rest go in. Everything entering a mapped source is mapped \u2014 jobs, syncs and `add()` alike \u2014 so `add()` on a mapped source returns `documents` (possibly several, with `invalid` naming any that failed) and throws `mapper_skipped` for an object the mapper refused; an object that maps to more than a thousand documents registers the first thousand and returns `job`, which takes the rest (`documentCount` is the total). A mapper may call models (`runTask`) or `fetch`; every call is spend per object. The owner tests one with `remy-admin datasources map test --source <slug> --dev` before deploying, reads what a job skipped with `datasources jobs quarantine <id>`, and re-applies a fixed mapper with `datasources remap --source <slug>`. See the dataSources skill for the authoring loop.\n\nTo write what each document says into the app\'s own tables (the equipment a passage names, the relations it states, the people it quotes), give the source a READER: `datasources/<slug>.reader.ts` exporting `defineReader(Source, { read, removed, cleared? })` from `@mindstudio-ai/agent`, declared in mindstudio.json as `"dataSources": [{ "slug": "<slug>", "reader": { "path": "dist/datasources/<slug>.reader.ts", "timeoutMs"?: 60000, "concurrency"?: 16 } }]` (beside or instead of a mapper). The platform runs it over every document the source builds, however it arrived (`add()`, a sync or a job), with its own batching, retries, budgets and pauses: `read(document)` gets `{ id, sourceId, filename, contentType, externalId, metadata, entities, pageCount, chunkCount, createdAt, url, chunks(), bytes() }`, where `chunks()` returns the passages as indexed, each with its entities, and `bytes()` (or the short-lived `url`) the file the document came from, for a reader that needs the drawing itself; `removed(documentId)` runs when a document leaves the source; `cleared(sourceId)` once when the source is deleted; the optional `relinked(document)` when a new dictionary re-tagged a document already read (the same document, with its new entities; for rows derived from the tags, not a reason to extract again). A document can be read more than once, so `read` replaces that document\'s rows as a set, keyed by its id, and `removed` deletes them. A throw fails that document and it is retried. The reader always runs against the LIVE release (its code and its database) and starts reading once the app publishes it; it may call models and `fetch`, and every call is spend per document. A reader may not add to, remove from or push a dictionary to its own source. The owner tests one with `remy-admin datasources reader test --source <slug> --dev`, sees its deliveries with `datasources reader status`, and reads what it hasn\'t with `datasources reread --source <slug> --unread`, a job priced before it runs. See the dataSources skill for the authoring loop.\n\nSupported: pdf, docx, pptx, xlsx, odt, rtf, epub, images, txt, md, json, csv, tsv, log, html.\n\nAdding the same bytes twice is free \u2014 documents are content-addressed, so a re-add is a no-op when this source has already processed those exact bytes under its current configuration. Only reconfiguring the source makes a re-add do work again, and that is an explicit owner-triggered migration, never a side effect of a deploy.\n\nEverything in `SearchOptions` \u2014 `mode`, `filter`, `rerank`, `hybrid`, `maxPerDocument`, `highlight` \u2014 is free to change and takes effect on the next search. How documents are chunked and embedded is configured on the corpus itself with `remy-admin datasources config` \u2014 not in code \u2014 and changing it on a populated source is rejected rather than silently rebuilding.\n\n**A data source is live and shared.** Unlike database tables there is no dev copy and no per-release isolation: adding or removing a document changes what the deployed app retrieves, immediately. Scenarios never reset a data source, and re-ingesting a large corpus costs real money.\n\n#### `analytics` \u2014 read the app\'s own traffic + event analytics\n\nThe platform auto-tracks pageviews (and custom events via the frontend\'s `analytics.track()`); the backend `analytics` namespace is the read side, so app code can correlate real traffic with its own data \u2014 lifetime views next to each blog post, "N people reading now", conversions beside the records that produced them. Backend only (hook token); reads the LIVE app\'s real traffic even from a dev sandbox. Rate limited per app (600 reads/min; the SDK auto-retries one 429). For pages composed of several reads, use `analytics.batch([spec, ...])` \u2014 up to 10 queries in one round trip, one rate token each, results in request order \u2014 instead of a parallel burst of `query()` calls competing with each other. NEVER swallow `rate_limited` errors into empty data: a page silently rendering zeros is worse than surfacing a retry state.\n\nOne general read, `query()`, shaped like the Plausible/GA4 query APIs:\n\n```typescript\nimport { analytics } from \'@mindstudio-ai/agent\';\n\n// Lifetime pageviews + unique visitors per page \u2014 top 10.\nconst top = await analytics.query({\n  metrics: [\'pageviews\', \'visitors\'],\n  dimensions: [\'path\'],\n  dateRange: \'all\',\n  limit: 10,\n});\n// top.results[0] \u2192 { dimensions: { path: \'/post/hello\' }, metrics: { pageviews: 4210, visitors: 1830 } }\n\n// Daily views for one post, full history.\nconst series = await analytics.query({\n  metrics: [\'pageviews\'],\n  dimensions: [\'time\'],\n  granularity: \'day\',\n  filters: [[\'is\', \'path\', [\'/post/hello\']]],\n  dateRange: \'all\',\n});\n```\n\nGrammar: `metrics` from `pageviews | visitors | visits | events`; `dimensions` is at most ONE entity dimension (`path`, `referrerHost`, `sourceCategory`, `country`, `city`, `deviceType`, `browser`, `os`, `language`, `visitorType`, `utmSource`\u2026`utmContent`, `eventName`) OR `\'time\'` (with `granularity: \'5m\'|\'hour\'|\'day\'|\'week\'|\'month\'` and optional IANA `timezone`); `filters` are `[op, dimension, values]` triples with ops `is | is_not | contains`; `dateRange` is a shorthand (`\'1h\'|\'24h\'|\'7d\'|\'30d\'|\'90d\'|\'all\'`) or `[startISO, endISO]`; `orderBy` (grouped only) accepts `pageviews` or `events`. With `dateRange: \'all\'`, granularity is a minimum \u2014 the server coarsens as history grows. `city` filter values are plain city names (any country) and always read the 90-day event detail.\n\n**Retention split \u2014 the one thing to get right:** queries whose filters are all `is` and touch at most one dimension read a rollup kept forever (full lifetime history, exact unique visitors). Cross-dimension intersections, `is_not`, and `contains` scan raw events retained for 90 days; the server clamps the window and reports it. Check `meta`: `source` (`rollup` vs `events`), `clamped`, `window.served`, and `metricsOmitted` \u2014 metrics a shape can\'t carry (e.g. `visits` on grouped reads) are omitted there, never returned as silent zeros. `meta.total` carries the unpaginated group count for paging.\n\nSpecials (same `dateRange`/filter-triple grammar): `analytics.live()` \u2192 `{count, countries, sparkline}` right now; `analytics.sources(opts?)` \u2192 per-session first-source ranking (UTM > referrer > direct) classified by category/vendor incl. AI assistants; `analytics.map(opts?)` \u2192 city lat/lon points; `analytics.aiSources(opts?)` \u2192 AI-assistant referrals per vendor; `analytics.crawlers.overview()/timeseries()/recent()` \u2192 bot/AI-crawler ingestion of the app\'s pages.\n\n#### `events` \u2014 realtime over named channels (publish/subscribe)\n\nPublish to named channels from any backend code (a method, a cron, a webhook handler); connected clients receive the payloads instantly over a platform-held stream \u2014 live dashboards, notifications, chat, collaborative UIs, multi-tab/device sync, with no polling and no WebSocket code. Different from `stream()`, which narrates ONE invocation to the caller currently waiting on it: events reach clients that were not part of the invocation at all. Backend only (hook token).\n\n```typescript\nimport { auth, events } from \'@mindstudio-ai/agent\';\n\n// The subscribe door is one of YOUR methods \u2014 auth checks first, then mint.\n// The grant is the ENTIRE subscribe-side authorization: whoever holds the\n// token receives those channels\' events for its lifetime.\nexport async function watchInbox() {\n  auth.requireRole(\'member\');\n  return await events.grant(`user:${auth.userId}`);  // { token, expiresAt, ttlSeconds }; ttlSeconds? 60\u20133600, default 900\n}\n\n// Anything can publish \u2014 one call, up to 500 channels:\nconst members = await RoomMembers.where({ roomId });\nconst { delivered } = await events.publish(\n  members.map((m) => `user:${m.userId}`),\n  { type: \'message\', roomId, msgId },\n);\n```\n\nThe frontend consumes with `events.connect({ getToken, onEvent, onConnect })` from `@mindstudio-ai/interface` \u2014 `getToken` calls the minting method, and the SDK handles reconnection and grant renewal.\n\n**Ephemeral client signals (cursors, typing, live strokes):** mint the grant with `publish` channels \u2014 `events.grant(channel, { publish: [channel] })` \u2014 and the frontend publishes DIRECTLY via `sub.publish(channels, data)`: no backend method runs per signal, the SDK coalesces to ~40ms batches, and failures drop (never retry). Client events cap at 8k serialized, 30 batches/sec per grant. Carry a client `seq` in the payload (delivery is unordered across batches; receivers drop stale seqs and interpolate). Durable state still commits through a method with an id-only nudge \u2014 signal ephemerally, commit through code.\n\n**A channel is an audience, and a method decides who is in it.** Never put two users\' data on one channel. Default to per-user channels with publish-time fan-out (as above) for membership-gated audiences \u2014 removing someone stops their events immediately; reserve a shared channel for genuinely broadcast content (a ticker, a live blog), where revocation waits out the grant TTL. For anonymous visitors `auth.userId` is null \u2014 key their channels on `session.visitorId` instead, or `user:null` becomes one channel shared by every anonymous user.\n\n**Events are at-most-once nudges.** Nothing is buffered while a client is disconnected and nothing replays on connect \u2014 clients refetch state in `onConnect` (subscribe for speed, reconcile for truth). `delivered: 0` means nobody is listening right now: normal, never an error. Every delivered frame carries the platform-stamped publish `id` (consumer dedupe key: `id` + `channel`); `publish` returns the same id. Payloads cap at 256k serialized chars, checked in the SDK before the network so oversize throws locally \u2014 when a publish carries data, publish before committing the write it announces; high-rate paths publish ids (`{ type, id }`) and let the client fetch. Channels are exact strings (letters, digits, `: _ - .` \u2014 no wildcards), \u2264500 per publish, \u2264100 per grant. Publishes and grants are scoped to the execution\'s environment (live/preview/dev) automatically, so a dev-session publish can never reach live users. Debug with `remy-admin events tail | publish | channels list`.\n';
   }
 });
 
@@ -9177,10 +8953,11 @@ async function runAsk(question, options = {}, onEvent) {
   });
   const system = await buildSystemPrompt(agent);
   const messages = [{ role: "user", content: question }];
+  let emptyRetries = 0;
   while (true) {
     let assistantText = "";
     const toolCalls = [];
-    let stopReason = "end_turn";
+    let stopReason = null;
     for await (const event of streamChat({
       baseUrl,
       apiKey,
@@ -9212,6 +8989,20 @@ async function runAsk(question, options = {}, onEvent) {
           throw new Error(event.error);
       }
     }
+    if (stopReason === null) {
+      throw new Error(
+        "The answer stream ended before the model finished. Ask again."
+      );
+    }
+    if (!assistantText.trim() && toolCalls.length === 0) {
+      if (emptyRetries < EMPTY_TURN_RETRIES) {
+        emptyRetries += 1;
+        continue;
+      }
+      throw new Error(
+        `The model returned an empty answer (stop reason: ${stopReason}). Ask again, or split the question.`
+      );
+    }
     messages.push({
       role: "assistant",
       content: assistantText,
@@ -9285,7 +9076,7 @@ async function cmdAsk(question, options) {
     process.exit(1);
   }
 }
-var DEFAULT_BASE_URL2, color, ansi;
+var DEFAULT_BASE_URL2, EMPTY_TURN_RETRIES, color, ansi;
 var init_ask = __esm({
   async "src/ask/index.ts"() {
     "use strict";
@@ -9295,6 +9086,7 @@ var init_ask = __esm({
     init_tools();
     init_prompt();
     DEFAULT_BASE_URL2 = "https://v1.mindstudio-api.com";
+    EMPTY_TURN_RETRIES = 1;
     color = process.stderr.isTTY;
     ansi = {
       dim: (s) => color ? `\x1B[2m${s}\x1B[0m` : s,
@@ -9364,7 +9156,7 @@ async function startMcpServer(options) {
           capabilities: { tools: {} },
           serverInfo: {
             name: "mindstudio-agent",
-            version: "0.1.139"
+            version: "0.1.141"
           },
           instructions: 'Welcome to MindStudio \u2014 a platform with 200+ AI models, 850+ third-party integrations, and pre-built agents.\n\nGetting started:\n1. Call `ask` with any question about the SDK \u2014 it knows every action, model, and connector and returns working code with real model IDs and config options. Examples: ask("generate an image with FLUX"), ask("what models support vision?"), ask("how do I send a Slack message?").\n2. Call `changeName` to set your display name \u2014 use your name or whatever your user calls you. This is how you\'ll appear in MindStudio request logs.\n3. If you have a profile picture, pass its public image URL to `changeProfilePicture`.\n4. For manual browsing, call `listActions` to discover all available actions.\n\nThen use the tools to generate text, images, video, audio, search the web, work with data sources, run agents, and more.\n\nImportant:\n- AI-powered actions (text generation, image generation, video, audio, etc.) cost money. Before running these, call `estimateActionCost` and confirm with the user before proceeding \u2014 unless they\'ve explicitly told you to go ahead.\n- Not all agents from `listAgents` are configured for API use. Do not try to run an agent just because it appears in the list \u2014 it will likely fail. Only run agents the user specifically asks you to run.'
         });
@@ -10376,7 +10168,7 @@ function isNewerVersion(current, latest) {
   return false;
 }
 async function checkForUpdate() {
-  const currentVersion = "0.1.139";
+  const currentVersion = "0.1.141";
   if (!currentVersion || process.env.MINDSTUDIO_NO_UPDATE_CHECK) {
     return null;
   }
@@ -10411,7 +10203,7 @@ async function checkForUpdate() {
   }
 }
 function printUpdateNotice(latestVersion) {
-  const currentVersion = "0.1.139";
+  const currentVersion = "0.1.141";
   process.stderr.write(
     `
   ${ansi2.cyanBright("Update available")} ${ansi2.gray(currentVersion + " \u2192")} ${ansi2.cyanBold(latestVersion)}
@@ -10424,7 +10216,7 @@ function isStandaloneBinary() {
   return !argv1.includes("node_modules");
 }
 async function cmdUpdate() {
-  const currentVersion = "0.1.139";
+  const currentVersion = "0.1.141";
   process.stderr.write(
     `  ${ansi2.gray("Current version:")} ${currentVersion}
 `
@@ -10565,7 +10357,7 @@ async function cmdLogin(options) {
   process.stderr.write("\n");
   printLogo();
   process.stderr.write("\n");
-  const ver = "0.1.139";
+  const ver = "0.1.141";
   process.stderr.write(
     `  ${ansi2.bold("MindStudio Agent")} ${ver ? " " + ansi2.gray("v" + ver) : ""}
 `
@@ -10893,7 +10685,7 @@ async function main() {
   const updatePromise = command !== "mcp" && command !== "login" && command !== "update" ? checkForUpdate() : Promise.resolve(null);
   try {
     if (command === "version" || command === "-v") {
-      process.stdout.write("0.1.139\n");
+      process.stdout.write("0.1.141\n");
       return;
     }
     if (command === "login") {