@modelrelay/sdk 1.3.1 → 1.3.3

package/README.md

@@ -1,220 +1,40 @@
 # ModelRelay TypeScript SDK
 
-```bash
-bun add @modelrelay/sdk
-```
-
-## Token Providers (Automatic Bearer Auth)
-
-Use token providers when you want the SDK to automatically obtain/refresh **bearer tokens** for data-plane calls like `/responses` and `/runs`.
-
-### OIDC id_token → customer bearer token (exchange)
-
-```ts
-import { ModelRelay, OIDCExchangeTokenProvider, parseSecretKey } from "@modelrelay/sdk";
-
-const tokenProvider = new OIDCExchangeTokenProvider({
-  apiKey: parseSecretKey(process.env.MODELRELAY_API_KEY!),
-  idTokenProvider: async () => {
-    // Return an OIDC id_token from your auth system (web login, device flow, etc).
-    return process.env.OIDC_ID_TOKEN!;
-  },
-});
-
-const mr = new ModelRelay({ tokenProvider });
-```
-
-If you need an `id_token` in a CLI-like context, you can use the OAuth device flow helper:
-
-```ts
-import { runOAuthDeviceFlowForIDToken } from "@modelrelay/sdk";
-
-const idToken = await runOAuthDeviceFlowForIDToken({
-  deviceAuthorizationEndpoint: "https://issuer.example.com/oauth/device/code",
-  tokenEndpoint: "https://issuer.example.com/oauth/token",
-  clientId: "your-client-id",
-  scope: "openid email profile",
-  onUserCode: ({ verificationUri, userCode }) => {
-    console.log(`Open ${verificationUri} and enter code: ${userCode}`);
-  },
-});
-```
-
-### Secret key → customer bearer token (mint)
-
-```ts
-import { CustomerTokenProvider, ModelRelay } from "@modelrelay/sdk";
-
-const tokenProvider = new CustomerTokenProvider({
-  secretKey: process.env.MODELRELAY_API_KEY!,
-  request: { projectId: "proj_...", customerId: "cust_..." },
-});
+The ModelRelay TypeScript SDK is a **responses-first**, **streaming-first** client for building cross-provider LLM features without committing to any single vendor API.
 
-const mr = new ModelRelay({ tokenProvider });
-```
-
-## Streaming Responses
-
-```ts
-import { ModelRelay } from "@modelrelay/sdk";
-
-const mr = ModelRelay.fromSecretKey("mr_sk_...");
-
-const req = mr.responses
-  .new()
-  .model("claude-sonnet-4-20250514")
-  .user("Hello")
-  .build();
-
-const stream = await mr.responses.stream(req);
+It's designed to feel great in TypeScript:
+- One fluent builder for **streaming/non-streaming**, **text/structured**, and **customer-attributed** requests
+- Structured outputs powered by Zod schemas with validation and retry
+- A practical tool-use toolkit for "LLM + tools" apps
 
-for await (const event of stream) {
-  if (event.type === "message_delta" && event.textDelta) {
-    process.stdout.write(event.textDelta);
-  }
-}
+```bash
+npm install @modelrelay/sdk
+# or
+bun add @modelrelay/sdk
 ```
 
-## Customer-Scoped Convenience
+## Quick Start
 
 ```ts
 import { ModelRelay } from "@modelrelay/sdk";
 
-const mr = ModelRelay.fromSecretKey("mr_sk_...");
-const customer = mr.forCustomer("cust_abc123");
+const mr = ModelRelay.fromSecretKey(process.env.MODELRELAY_API_KEY!);
 
-const text = await customer.responses.text(
-  "You are a helpful assistant.",
-  "Summarize Q4 results",
+const response = await mr.responses.create(
+  mr.responses
+    .new()
+    .model("claude-sonnet-4-20250514")
+    .system("Answer concisely.")
+    .user("Write one line about TypeScript.")
+    .build()
 );
-```
-
-You can also stream structured JSON for a specific customer:
-
-```ts
-import { z } from "zod";
-import { ModelRelay, outputFormatFromZod } from "@modelrelay/sdk";
 
-const mr = ModelRelay.fromSecretKey("mr_sk_...");
-const customer = mr.forCustomer("cust_abc123");
-
-const schema = z.object({
-  summary: z.string(),
-  highlights: z.array(z.string()),
-});
-
-const req = customer.responses
-  .new()
-  .outputFormat(outputFormatFromZod(schema))
-  .system("You are a helpful assistant.")
-  .user("Summarize Q4 results")
-  .build();
-
-const stream = await customer.responses.streamJSON<z.infer<typeof schema>>(req);
-for await (const event of stream) {
-  if (event.type === "completion") {
-    console.log(event.value);
-  }
-}
-```
-
-You can also pass a single object to `textForCustomer`:
-
-```ts
-const text = await mr.responses.textForCustomer({
-  customerId: "cust_abc123",
-  system: "You are a helpful assistant.",
-  user: "Summarize Q4 results",
-});
+console.log(response.text());
 ```
 
-## Workflow Runs (workflow.v0)
-
-```ts
-import {
-  ModelRelay,
-  type LLMResponsesBindingV0,
-  parseNodeId,
-  parseOutputName,
-  parseSecretKey,
-  workflowV0,
-} from "@modelrelay/sdk";
-
-const mr = new ModelRelay({ key: parseSecretKey("mr_sk_...") });
-
-const spec = workflowV0()
-  .name("multi_agent_v0_example")
-  .execution({ max_parallelism: 3, node_timeout_ms: 20_000, run_timeout_ms: 30_000 })
-  .llmResponses(parseNodeId("agent_a"), {
-    model: "claude-sonnet-4-20250514",
-    input: [
-      { type: "message", role: "system", content: [{ type: "text", text: "You are Agent A." }] },
-      { type: "message", role: "user", content: [{ type: "text", text: "Write 3 ideas for a landing page." }] },
-    ],
-  })
-  .llmResponses(parseNodeId("agent_b"), {
-    model: "claude-sonnet-4-20250514",
-    input: [
-      { type: "message", role: "system", content: [{ type: "text", text: "You are Agent B." }] },
-      { type: "message", role: "user", content: [{ type: "text", text: "Write 3 objections a user might have." }] },
-    ],
-  })
-  .llmResponses(parseNodeId("agent_c"), {
-    model: "claude-sonnet-4-20250514",
-    input: [
-      { type: "message", role: "system", content: [{ type: "text", text: "You are Agent C." }] },
-      { type: "message", role: "user", content: [{ type: "text", text: "Write 3 alternative headlines." }] },
-    ],
-  })
-  .joinAll(parseNodeId("join"))
-  .llmResponses(
-    parseNodeId("aggregate"),
-    {
-      model: "claude-sonnet-4-20250514",
-      input: [
-        {
-          type: "message",
-          role: "system",
-          content: [{ type: "text", text: "Synthesize the best answer from the following agent outputs (JSON)." }],
-        },
-        { type: "message", role: "user", content: [{ type: "text", text: "" }] }, // overwritten by bindings
-      ],
-    },
-    {
-      // Bind the join output into the aggregator prompt (fan-in).
-      bindings: [
-        {
-          from: parseNodeId("join"),
-          to: "/input/1/content/0/text",
-          encoding: "json_string",
-        } satisfies LLMResponsesBindingV0,
-      ],
-    },
-  )
-  .edge(parseNodeId("agent_a"), parseNodeId("join"))
-  .edge(parseNodeId("agent_b"), parseNodeId("join"))
-  .edge(parseNodeId("agent_c"), parseNodeId("join"))
-  .edge(parseNodeId("join"), parseNodeId("aggregate"))
-  .output(parseOutputName("result"), parseNodeId("aggregate"))
-  .build();
-
-const { run_id } = await mr.runs.create(spec);
-
-const events = await mr.runs.events(run_id);
-for await (const ev of events) {
-  if (ev.type === "run_completed") {
-    const status = await mr.runs.get(run_id);
-    console.log("outputs:", status.outputs);
-    console.log("cost_summary:", status.cost_summary);
-  }
-}
-```
-
-See the full example in `sdk/ts/examples/workflows_multi_agent.ts`.
-
 ## Chat-Like Text Helpers
 
-For the most common path (**system + user → assistant text**):
+For the most common path (**system + user → assistant text**), use the built-in convenience:
 
 ```ts
 const text = await mr.responses.text(
@@ -228,14 +48,16 @@ console.log(text);
 For customer-attributed requests where the backend selects the model:
 
 ```ts
-const text = await mr.responses.textForCustomer(
-  "customer-123",
-  "Answer concisely.",
-  "Say hi.",
+const customer = mr.forCustomer("customer-123");
+const text = await customer.responses.text(
+  "You are a helpful assistant.",
+  "Summarize Q4 results",
 );
 ```
 
-To stream only message text deltas:
+## Streaming
+
+Stream text deltas for real-time output:
 
 ```ts
 const deltas = await mr.responses.streamTextDeltas(
@@ -248,14 +70,31 @@ for await (const delta of deltas) {
 }
 ```
 
+For full control, stream typed events:
+
+```ts
+const req = mr.responses
+  .new()
+  .model("claude-sonnet-4-20250514")
+  .user("Hello")
+  .build();
+
+const stream = await mr.responses.stream(req);
+
+for await (const event of stream) {
+  if (event.type === "message_delta" && event.textDelta) {
+    process.stdout.write(event.textDelta);
+  }
+}
+```
+
 ## Structured Outputs with Zod
 
+Get typed, validated responses from the model:
+
 ```ts
-import { ModelRelay, parseSecretKey } from "@modelrelay/sdk";
 import { z } from "zod";
 
-const mr = new ModelRelay({ key: parseSecretKey("mr_sk_...") });
-
 const Person = z.object({
   name: z.string(),
   age: z.number(),
@@ -270,16 +109,11 @@ const result = await mr.responses.structured(
 console.log(result.value); // { name: "John Doe", age: 30 }
 ```
 
-## Streaming Structured Outputs
+### Streaming Structured Outputs
 
 Build progressive UIs that render fields as they complete:
 
 ```ts
-import { ModelRelay, parseSecretKey } from "@modelrelay/sdk";
-import { z } from "zod";
-
-const mr = new ModelRelay({ key: parseSecretKey("mr_sk_...") });
-
 const Article = z.object({
   title: z.string(),
   summary: z.string(),
@@ -292,15 +126,12 @@ const stream = await mr.responses.streamStructured(
 );
 
 for await (const event of stream) {
-  // Render fields as soon as they're complete
   if (event.completeFields.has("title")) {
     renderTitle(event.payload.title);  // Safe to display
   }
   if (event.completeFields.has("summary")) {
     renderSummary(event.payload.summary);
   }
-
-  // Show streaming preview of incomplete fields
   if (!event.completeFields.has("body")) {
     renderBodyPreview(event.payload.body + "▋");
   }
@@ -309,7 +140,7 @@ for await (const event of stream) {
 
 ## Customer-Attributed Requests
 
-For metered billing, use `customerId()` — the customer's tier determines the model and `model` can be omitted:
+For metered billing, use `customerId()`. The customer's tier determines the model, so `model()` can be omitted:
 
 ```ts
 const req = mr.responses
@@ -321,6 +152,16 @@ const req = mr.responses
 const stream = await mr.responses.stream(req);
 ```
 
+Or use the convenience method:
+
+```ts
+const text = await mr.responses.textForCustomer(
+  "customer-123",
+  "Answer concisely.",
+  "Say hi.",
+);
+```
+
 ## Customer Management (Backend)
 
 ```ts
@@ -341,6 +182,92 @@ const session = await mr.customers.createCheckoutSession(customer.id, {
 const status = await mr.customers.getSubscription(customer.id);
 ```
 
+## Workflow Runs
+
+Build multi-agent workflows with parallel execution:
+
+```ts
+import { workflowV0, parseNodeId, parseOutputName, type LLMResponsesBindingV0 } from "@modelrelay/sdk";
+
+const spec = workflowV0()
+  .name("multi_agent_example")
+  .execution({ max_parallelism: 3, node_timeout_ms: 20_000, run_timeout_ms: 30_000 })
+  .llmResponses(parseNodeId("agent_a"), {
+    model: "claude-sonnet-4-20250514",
+    input: [
+      { type: "message", role: "system", content: [{ type: "text", text: "You are Agent A." }] },
+      { type: "message", role: "user", content: [{ type: "text", text: "Analyze the question." }] },
+    ],
+  })
+  .llmResponses(parseNodeId("agent_b"), {
+    model: "claude-sonnet-4-20250514",
+    input: [
+      { type: "message", role: "system", content: [{ type: "text", text: "You are Agent B." }] },
+      { type: "message", role: "user", content: [{ type: "text", text: "Find edge cases." }] },
+    ],
+  })
+  .joinAll(parseNodeId("join"))
+  .llmResponses(
+    parseNodeId("aggregate"),
+    {
+      model: "claude-sonnet-4-20250514",
+      input: [
+        { type: "message", role: "system", content: [{ type: "text", text: "Synthesize the best answer." }] },
+        { type: "message", role: "user", content: [{ type: "text", text: "" }] },
+      ],
+    },
+    {
+      bindings: [
+        { from: parseNodeId("join"), to: "/input/1/content/0/text", encoding: "json_string" } satisfies LLMResponsesBindingV0,
+      ],
+    },
+  )
+  .edge(parseNodeId("agent_a"), parseNodeId("join"))
+  .edge(parseNodeId("agent_b"), parseNodeId("join"))
+  .edge(parseNodeId("join"), parseNodeId("aggregate"))
+  .output(parseOutputName("result"), parseNodeId("aggregate"))
+  .build();
+
+const { run_id } = await mr.runs.create(spec);
+
+for await (const ev of await mr.runs.events(run_id)) {
+  if (ev.type === "run_completed") {
+    const status = await mr.runs.get(run_id);
+    console.log("outputs:", status.outputs);
+  }
+}
+```
+
+## Token Providers (Advanced)
+
+For automatic bearer token management in data-plane calls:
+
+### Secret key → customer bearer token
+
+```ts
+import { CustomerTokenProvider, ModelRelay } from "@modelrelay/sdk";
+
+const tokenProvider = new CustomerTokenProvider({
+  secretKey: process.env.MODELRELAY_API_KEY!,
+  request: { projectId: "proj_...", customerId: "cust_..." },
+});
+
+const mr = new ModelRelay({ tokenProvider });
+```
+
+### OIDC exchange
+
+```ts
+import { ModelRelay, OIDCExchangeTokenProvider, parseSecretKey } from "@modelrelay/sdk";
+
+const tokenProvider = new OIDCExchangeTokenProvider({
+  apiKey: parseSecretKey(process.env.MODELRELAY_API_KEY!),
+  idTokenProvider: async () => process.env.OIDC_ID_TOKEN!,
+});
+
+const mr = new ModelRelay({ tokenProvider });
+```
+
 ## Configuration
 
 ```ts

package/dist/index.cjs

@@ -818,7 +818,7 @@ function isTokenReusable(token) {
 // package.json
 var package_default = {
   name: "@modelrelay/sdk",
-  version: "1.3.1",
+  version: "1.3.3",
   description: "TypeScript SDK for the ModelRelay API",
   type: "module",
   main: "dist/index.cjs",
@@ -850,6 +850,9 @@ var package_default = {
   ],
   author: "Shane Vitarana",
   license: "Apache-2.0",
+  dependencies: {
+    "fast-json-patch": "^3.1.1"
+  },
   devDependencies: {
     "openapi-typescript": "^7.4.4",
     tsup: "^8.2.4",
@@ -2042,8 +2045,11 @@ function mapNDJSONResponseEvent(line, requestId) {
   const model = normalizeModelId(parsed.model);
   const stopReason = normalizeStopReason(parsed.stop_reason);
   let textDelta;
-  if (isRecord(parsed.payload) && typeof parsed.payload.content === "string") {
-    textDelta = parsed.payload.content;
+  if (recordType2 === "update" && typeof parsed.delta === "string") {
+    textDelta = parsed.delta;
+  }
+  if (recordType2 === "completion" && typeof parsed.content === "string") {
+    textDelta = parsed.content;
   }
   const toolCallDelta = extractToolCallDelta(parsed, type);
   const toolCalls = extractToolCalls(parsed, type);
@@ -2143,6 +2149,7 @@ function normalizeToolCalls(toolCalls) {
 }
 
 // src/responses_stream.ts
+var import_fast_json_patch = require("fast-json-patch");
 var ResponsesStream = class {
   constructor(response, requestId, context, metrics, trace, timeouts = {}, startedAtMs = Date.now()) {
     this.firstTokenEmitted = false;
@@ -2269,6 +2276,9 @@ var ResponsesStream = class {
       if (evt.type === "message_stop") {
         stopReason = evt.stopReason;
         usage = evt.usage;
+        if (evt.textDelta) {
+          text = evt.textDelta;
+        }
         const raw2 = isRecord(evt.data) ? evt.data : {};
         citations = normalizeCitations(raw2.citations) ?? citations;
       }
@@ -2344,6 +2354,7 @@ var StructuredJSONStream = class {
     this.closed = false;
     this.sawTerminal = false;
     this.firstContentSeen = false;
+    this.currentPayload = {};
     if (!response.body) {
       throw new ConfigError("streaming response is missing a body");
     }
@@ -2487,9 +2498,36 @@ var StructuredJSONStream = class {
     }
     this.firstContentSeen = true;
     const completeFieldsArray = Array.isArray(parsed.complete_fields) ? parsed.complete_fields.filter((f) => typeof f === "string") : [];
+    let payload;
+    if (rawType === "update") {
+      const patch = Array.isArray(parsed.patch) ? parsed.patch : null;
+      if (!patch) {
+        throw new TransportError("structured stream update missing patch", {
+          kind: "request"
+        });
+      }
+      try {
+        const applied = (0, import_fast_json_patch.applyPatch)(this.currentPayload, patch, false, true);
+        this.currentPayload = applied.newDocument;
+        payload = this.currentPayload;
+      } catch (err) {
+        throw new TransportError("failed to apply structured patch", {
+          kind: "request",
+          cause: err
+        });
+      }
+    } else {
+      if (parsed.payload === void 0) {
+        throw new TransportError("structured stream completion missing payload", {
+          kind: "request"
+        });
+      }
+      this.currentPayload = parsed.payload;
+      payload = this.currentPayload;
+    }
     return {
       type: rawType,
-      payload: parsed.payload,
+      payload,
       requestId: this.requestId,
       completeFields: new Set(completeFieldsArray)
     };
@@ -2599,6 +2637,7 @@ async function readWithTimeout(readPromise, next, onTimeout) {
 }
 
 // src/responses_client.ts
+var RESPONSES_STREAM_ACCEPT = 'application/x-ndjson; profile="responses-stream/v2"';
 var ResponsesClient = class {
   constructor(http, auth, cfg = {}) {
     this.http = http;
@@ -2638,24 +2677,16 @@ var ResponsesClient = class {
     const stream = await this.stream(req, options);
     return {
       async *[Symbol.asyncIterator]() {
-        let accumulated = "";
+        let sawDelta = false;
         try {
           for await (const evt of stream) {
-            if ((evt.type === "message_delta" || evt.type === "message_stop") && evt.textDelta) {
-              const next = evt.textDelta;
-              let delta = "";
-              if (next.startsWith(accumulated)) {
-                delta = next.slice(accumulated.length);
-                accumulated = next;
-              } else if (accumulated.startsWith(next)) {
-                accumulated = next;
-              } else {
-                delta = next;
-                accumulated += next;
-              }
-              if (delta) {
-                yield delta;
-              }
+            if (evt.type === "message_delta" && evt.textDelta) {
+              sawDelta = true;
+              yield evt.textDelta;
+            }
+            if (evt.type === "message_stop" && evt.textDelta && !sawDelta) {
+              sawDelta = true;
+              yield evt.textDelta;
             }
           }
         } finally {
@@ -2673,24 +2704,16 @@ var ResponsesClient = class {
     const stream = await this.stream(req, args.options ?? options);
     return {
       async *[Symbol.asyncIterator]() {
-        let accumulated = "";
+        let sawDelta = false;
         try {
           for await (const evt of stream) {
-            if ((evt.type === "message_delta" || evt.type === "message_stop") && evt.textDelta) {
-              const next = evt.textDelta;
-              let delta = "";
-              if (next.startsWith(accumulated)) {
-                delta = next.slice(accumulated.length);
-                accumulated = next;
-              } else if (accumulated.startsWith(next)) {
-                accumulated = next;
-              } else {
-                delta = next;
-                accumulated += next;
-              }
-              if (delta) {
-                yield delta;
-              }
+            if (evt.type === "message_delta" && evt.textDelta) {
+              sawDelta = true;
+              yield evt.textDelta;
+            }
+            if (evt.type === "message_stop" && evt.textDelta && !sawDelta) {
+              sawDelta = true;
+              yield evt.textDelta;
             }
           }
         } finally {
@@ -2789,7 +2812,7 @@ var ResponsesClient = class {
       headers,
       apiKey: authHeaders.apiKey,
       accessToken: authHeaders.accessToken,
-      accept: "application/x-ndjson",
+      accept: RESPONSES_STREAM_ACCEPT,
       raw: true,
       signal: merged.signal,
       timeoutMs: merged.timeoutMs ?? 0,
@@ -2869,7 +2892,7 @@ var ResponsesClient = class {
       headers,
       apiKey: authHeaders.apiKey,
       accessToken: authHeaders.accessToken,
-      accept: "application/x-ndjson",
+      accept: RESPONSES_STREAM_ACCEPT,
       raw: true,
       signal: merged.signal,
       timeoutMs: merged.timeoutMs ?? 0,

package/dist/index.d.cts

@@ -1668,6 +1668,7 @@ declare class StructuredJSONStream<T> implements AsyncIterable<StructuredJSONEve
     private closed;
     private sawTerminal;
     private firstContentSeen;
+    private currentPayload;
     constructor(response: globalThis.Response, requestId: string | undefined, context: RequestContext, _metrics?: MetricsCallbacks, trace?: TraceCallbacks, timeouts?: StreamTimeoutsMs, startedAtMs?: number);
     cancel(reason?: unknown): Promise<void>;
     [Symbol.asyncIterator](): AsyncIterator<StructuredJSONEvent<T>>;
@@ -3405,9 +3406,25 @@ interface components {
             /** @description Sources from web search results */
             citations?: components["schemas"]["Citation"][];
         };
+        JSONPatchOperation: {
+            /** @enum {string} */
+            op: "add" | "remove" | "replace" | "move" | "copy" | "test";
+            path: string;
+            from?: string;
+            value?: {
+                [key: string]: unknown;
+            };
+        };
         ResponsesStreamEnvelope: {
             /** @enum {string} */
             type: "start" | "update" | "completion" | "error" | "keepalive" | "tool_use_start" | "tool_use_delta" | "tool_use_stop";
+            /** @enum {string} */
+            stream_mode?: "text-delta" | "structured-patch";
+            /** @enum {string} */
+            stream_version?: "v2";
+            delta?: string;
+            content?: string;
+            patch?: components["schemas"]["JSONPatchOperation"][];
             payload?: {
                 [key: string]: unknown;
             };
@@ -3702,7 +3719,7 @@ interface components {
          * @description Workflow-critical model capability identifier.
          * @enum {string}
          */
-        ModelCapability: "tools" | "vision" | "web_search" | "computer_use" | "code_execution";
+        ModelCapability: "tools" | "vision" | "web_search" | "web_fetch" | "computer_use" | "code_execution";
         Model: {
             model_id: components["schemas"]["ModelId"];
             provider: components["schemas"]["ProviderId"];

package/dist/index.d.ts

@@ -1668,6 +1668,7 @@ declare class StructuredJSONStream<T> implements AsyncIterable<StructuredJSONEve
     private closed;
     private sawTerminal;
     private firstContentSeen;
+    private currentPayload;
     constructor(response: globalThis.Response, requestId: string | undefined, context: RequestContext, _metrics?: MetricsCallbacks, trace?: TraceCallbacks, timeouts?: StreamTimeoutsMs, startedAtMs?: number);
     cancel(reason?: unknown): Promise<void>;
     [Symbol.asyncIterator](): AsyncIterator<StructuredJSONEvent<T>>;
@@ -3405,9 +3406,25 @@ interface components {
             /** @description Sources from web search results */
             citations?: components["schemas"]["Citation"][];
         };
+        JSONPatchOperation: {
+            /** @enum {string} */
+            op: "add" | "remove" | "replace" | "move" | "copy" | "test";
+            path: string;
+            from?: string;
+            value?: {
+                [key: string]: unknown;
+            };
+        };
         ResponsesStreamEnvelope: {
             /** @enum {string} */
             type: "start" | "update" | "completion" | "error" | "keepalive" | "tool_use_start" | "tool_use_delta" | "tool_use_stop";
+            /** @enum {string} */
+            stream_mode?: "text-delta" | "structured-patch";
+            /** @enum {string} */
+            stream_version?: "v2";
+            delta?: string;
+            content?: string;
+            patch?: components["schemas"]["JSONPatchOperation"][];
             payload?: {
                 [key: string]: unknown;
             };
@@ -3702,7 +3719,7 @@ interface components {
          * @description Workflow-critical model capability identifier.
          * @enum {string}
          */
-        ModelCapability: "tools" | "vision" | "web_search" | "computer_use" | "code_execution";
+        ModelCapability: "tools" | "vision" | "web_search" | "web_fetch" | "computer_use" | "code_execution";
         Model: {
             model_id: components["schemas"]["ModelId"];
             provider: components["schemas"]["ProviderId"];

package/dist/index.js

@@ -688,7 +688,7 @@ function isTokenReusable(token) {
 // package.json
 var package_default = {
   name: "@modelrelay/sdk",
-  version: "1.3.1",
+  version: "1.3.3",
   description: "TypeScript SDK for the ModelRelay API",
   type: "module",
   main: "dist/index.cjs",
@@ -720,6 +720,9 @@ var package_default = {
   ],
   author: "Shane Vitarana",
   license: "Apache-2.0",
+  dependencies: {
+    "fast-json-patch": "^3.1.1"
+  },
   devDependencies: {
     "openapi-typescript": "^7.4.4",
     tsup: "^8.2.4",
@@ -1912,8 +1915,11 @@ function mapNDJSONResponseEvent(line, requestId) {
   const model = normalizeModelId(parsed.model);
   const stopReason = normalizeStopReason(parsed.stop_reason);
   let textDelta;
-  if (isRecord(parsed.payload) && typeof parsed.payload.content === "string") {
-    textDelta = parsed.payload.content;
+  if (recordType2 === "update" && typeof parsed.delta === "string") {
+    textDelta = parsed.delta;
+  }
+  if (recordType2 === "completion" && typeof parsed.content === "string") {
+    textDelta = parsed.content;
   }
   const toolCallDelta = extractToolCallDelta(parsed, type);
   const toolCalls = extractToolCalls(parsed, type);
@@ -2013,6 +2019,7 @@ function normalizeToolCalls(toolCalls) {
 }
 
 // src/responses_stream.ts
+import { applyPatch } from "fast-json-patch";
 var ResponsesStream = class {
   constructor(response, requestId, context, metrics, trace, timeouts = {}, startedAtMs = Date.now()) {
     this.firstTokenEmitted = false;
@@ -2139,6 +2146,9 @@ var ResponsesStream = class {
       if (evt.type === "message_stop") {
         stopReason = evt.stopReason;
         usage = evt.usage;
+        if (evt.textDelta) {
+          text = evt.textDelta;
+        }
         const raw2 = isRecord(evt.data) ? evt.data : {};
         citations = normalizeCitations(raw2.citations) ?? citations;
       }
@@ -2214,6 +2224,7 @@ var StructuredJSONStream = class {
     this.closed = false;
     this.sawTerminal = false;
     this.firstContentSeen = false;
+    this.currentPayload = {};
     if (!response.body) {
       throw new ConfigError("streaming response is missing a body");
     }
@@ -2357,9 +2368,36 @@ var StructuredJSONStream = class {
     }
     this.firstContentSeen = true;
     const completeFieldsArray = Array.isArray(parsed.complete_fields) ? parsed.complete_fields.filter((f) => typeof f === "string") : [];
+    let payload;
+    if (rawType === "update") {
+      const patch = Array.isArray(parsed.patch) ? parsed.patch : null;
+      if (!patch) {
+        throw new TransportError("structured stream update missing patch", {
+          kind: "request"
+        });
+      }
+      try {
+        const applied = applyPatch(this.currentPayload, patch, false, true);
+        this.currentPayload = applied.newDocument;
+        payload = this.currentPayload;
+      } catch (err) {
+        throw new TransportError("failed to apply structured patch", {
+          kind: "request",
+          cause: err
+        });
+      }
+    } else {
+      if (parsed.payload === void 0) {
+        throw new TransportError("structured stream completion missing payload", {
+          kind: "request"
+        });
+      }
+      this.currentPayload = parsed.payload;
+      payload = this.currentPayload;
+    }
     return {
       type: rawType,
-      payload: parsed.payload,
+      payload,
       requestId: this.requestId,
       completeFields: new Set(completeFieldsArray)
     };
@@ -2469,6 +2507,7 @@ async function readWithTimeout(readPromise, next, onTimeout) {
 }
 
 // src/responses_client.ts
+var RESPONSES_STREAM_ACCEPT = 'application/x-ndjson; profile="responses-stream/v2"';
 var ResponsesClient = class {
   constructor(http, auth, cfg = {}) {
     this.http = http;
@@ -2508,24 +2547,16 @@ var ResponsesClient = class {
     const stream = await this.stream(req, options);
     return {
       async *[Symbol.asyncIterator]() {
-        let accumulated = "";
+        let sawDelta = false;
         try {
           for await (const evt of stream) {
-            if ((evt.type === "message_delta" || evt.type === "message_stop") && evt.textDelta) {
-              const next = evt.textDelta;
-              let delta = "";
-              if (next.startsWith(accumulated)) {
-                delta = next.slice(accumulated.length);
-                accumulated = next;
-              } else if (accumulated.startsWith(next)) {
-                accumulated = next;
-              } else {
-                delta = next;
-                accumulated += next;
-              }
-              if (delta) {
-                yield delta;
-              }
+            if (evt.type === "message_delta" && evt.textDelta) {
+              sawDelta = true;
+              yield evt.textDelta;
+            }
+            if (evt.type === "message_stop" && evt.textDelta && !sawDelta) {
+              sawDelta = true;
+              yield evt.textDelta;
             }
           }
         } finally {
@@ -2543,24 +2574,16 @@ var ResponsesClient = class {
     const stream = await this.stream(req, args.options ?? options);
     return {
       async *[Symbol.asyncIterator]() {
-        let accumulated = "";
+        let sawDelta = false;
         try {
           for await (const evt of stream) {
-            if ((evt.type === "message_delta" || evt.type === "message_stop") && evt.textDelta) {
-              const next = evt.textDelta;
-              let delta = "";
-              if (next.startsWith(accumulated)) {
-                delta = next.slice(accumulated.length);
-                accumulated = next;
-              } else if (accumulated.startsWith(next)) {
-                accumulated = next;
-              } else {
-                delta = next;
-                accumulated += next;
-              }
-              if (delta) {
-                yield delta;
-              }
+            if (evt.type === "message_delta" && evt.textDelta) {
+              sawDelta = true;
+              yield evt.textDelta;
+            }
+            if (evt.type === "message_stop" && evt.textDelta && !sawDelta) {
+              sawDelta = true;
+              yield evt.textDelta;
             }
           }
         } finally {
@@ -2659,7 +2682,7 @@ var ResponsesClient = class {
       headers,
       apiKey: authHeaders.apiKey,
       accessToken: authHeaders.accessToken,
-      accept: "application/x-ndjson",
+      accept: RESPONSES_STREAM_ACCEPT,
       raw: true,
       signal: merged.signal,
       timeoutMs: merged.timeoutMs ?? 0,
@@ -2739,7 +2762,7 @@ var ResponsesClient = class {
       headers,
       apiKey: authHeaders.apiKey,
       accessToken: authHeaders.accessToken,
-      accept: "application/x-ndjson",
+      accept: RESPONSES_STREAM_ACCEPT,
       raw: true,
       signal: merged.signal,
       timeoutMs: merged.timeoutMs ?? 0,

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@modelrelay/sdk",
-  "version": "1.3.1",
+  "version": "1.3.3",
 	"description": "TypeScript SDK for the ModelRelay API",
 	"type": "module",
 	"main": "dist/index.cjs",
@@ -32,6 +32,9 @@
 	],
 	"author": "Shane Vitarana",
 	"license": "Apache-2.0",
+	"dependencies": {
+		"fast-json-patch": "^3.1.1"
+	},
 	"devDependencies": {
 		"openapi-typescript": "^7.4.4",
 		"tsup": "^8.2.4",