mervexai 0.0.97

package/src/tool/tool.ts

@@ -0,0 +1,183 @@
+import { PermissionV1 } from "@msrfteam/core/v1/permission"
+import { Effect, Schema } from "effect"
+import { SessionV1 } from "@msrfteam/core/v1/session"
+import type { JSONSchema7 } from "@ai-sdk/provider"
+import type { MessageV2 } from "../session/message-v2"
+import type { Permission } from "../permission"
+import type { SessionID, MessageID } from "../session/schema"
+import * as Truncate from "./truncate"
+import { Agent } from "@/agent/agent"
+
+interface Metadata {
+  [key: string]: any
+}
+
+// TODO: remove this hack
+export type DynamicDescription = (agent: Agent.Info) => Effect.Effect<string>
+
+/**
+ * Raised when the LLM calls a tool with arguments that fail the parameter
+ * schema. This is the canonical "rewrite the input" tool error: the typed
+ * error class makes it matchable upstream, and its `message` getter produces
+ * the model-facing prose that the AI SDK feeds back as the tool result.
+ */
+export class InvalidArgumentsError extends Schema.TaggedErrorClass<InvalidArgumentsError>()(
+  "ToolInvalidArgumentsError",
+  {
+    tool: Schema.String,
+    detail: Schema.String,
+  },
+) {
+  override get message() {
+    return `The ${this.tool} tool was called with invalid arguments: ${this.detail}.\nPlease rewrite the input so it satisfies the expected schema.`
+  }
+}
+
+export type Context<M extends Metadata = Metadata> = {
+  sessionID: SessionID
+  messageID: MessageID
+  agent: string
+  abort: AbortSignal
+  callID?: string
+  extra?: { [key: string]: unknown }
+  messages: SessionV1.WithParts[]
+  metadata(input: { title?: string; metadata?: M }): Effect.Effect<void>
+  ask(input: Omit<PermissionV1.Request, "id" | "sessionID" | "tool">): Effect.Effect<void>
+}
+
+export interface ExecuteResult<M extends Metadata = Metadata> {
+  title: string
+  metadata: M
+  output: string
+  attachments?: Omit<SessionV1.FilePart, "id" | "sessionID" | "messageID">[]
+}
+
+export interface Def<
+  Parameters extends Schema.Decoder<unknown> = Schema.Decoder<unknown>,
+  M extends Metadata = Metadata,
+> {
+  id: string
+  description: string
+  parameters: Parameters
+  jsonSchema?: JSONSchema7
+  execute(args: Schema.Schema.Type<Parameters>, ctx: Context): Effect.Effect<ExecuteResult<M>>
+  formatValidationError?(error: unknown): string
+}
+export type DefWithoutID<
+  Parameters extends Schema.Decoder<unknown> = Schema.Decoder<unknown>,
+  M extends Metadata = Metadata,
+> = Omit<Def<Parameters, M>, "id">
+
+export interface Info<
+  Parameters extends Schema.Decoder<unknown> = Schema.Decoder<unknown>,
+  M extends Metadata = Metadata,
+> {
+  id: string
+  init: () => Effect.Effect<DefWithoutID<Parameters, M>>
+}
+
+type Init<Parameters extends Schema.Decoder<unknown>, M extends Metadata> =
+  | DefWithoutID<Parameters, M>
+  | (() => Effect.Effect<DefWithoutID<Parameters, M>>)
+
+export type InferParameters<T> =
+  T extends Info<infer P, any>
+    ? Schema.Schema.Type<P>
+    : T extends Effect.Effect<Info<infer P, any>, any, any>
+      ? Schema.Schema.Type<P>
+      : never
+export type InferMetadata<T> =
+  T extends Info<any, infer M> ? M : T extends Effect.Effect<Info<any, infer M>, any, any> ? M : never
+
+export type InferDef<T> =
+  T extends Info<infer P, infer M>
+    ? Def<P, M>
+    : T extends Effect.Effect<Info<infer P, infer M>, any, any>
+      ? Def<P, M>
+      : never
+
+function wrap<Parameters extends Schema.Decoder<unknown>, Result extends Metadata>(
+  id: string,
+  init: Init<Parameters, Result>,
+  truncate: Truncate.Interface,
+  agents: Agent.Interface,
+) {
+  return () =>
+    Effect.gen(function* () {
+      const toolInfo = typeof init === "function" ? { ...(yield* init()) } : { ...init }
+      // Compile the parser closure once per tool init; `decodeUnknownEffect`
+      // allocates a new closure per call, so hoisting avoids re-closing it for
+      // every LLM tool invocation.
+      const decode = Schema.decodeUnknownEffect(toolInfo.parameters)
+      const execute = toolInfo.execute
+      toolInfo.execute = (args, ctx) => {
+        const attrs = {
+          "tool.name": id,
+          "session.id": ctx.sessionID,
+          "message.id": ctx.messageID,
+          ...(ctx.callID ? { "tool.call_id": ctx.callID } : {}),
+        }
+        return Effect.gen(function* () {
+          const decoded = yield* decode(args).pipe(
+            Effect.mapError(
+              (error) =>
+                new InvalidArgumentsError({
+                  tool: id,
+                  detail: toolInfo.formatValidationError ? toolInfo.formatValidationError(error) : String(error),
+                }),
+            ),
+          )
+          const result = yield* execute(decoded as Schema.Schema.Type<Parameters>, ctx)
+          if (result.metadata.truncated !== undefined) {
+            return result
+          }
+          const agent = yield* agents.get(ctx.agent)
+          const truncated = yield* truncate.output(result.output, {}, agent)
+          return {
+            ...result,
+            output: truncated.content,
+            metadata: {
+              ...result.metadata,
+              truncated: truncated.truncated,
+              ...(truncated.truncated && { outputPath: truncated.outputPath }),
+            },
+          }
+        }).pipe(Effect.orDie, Effect.withSpan("Tool.execute", { attributes: attrs }))
+      }
+      return toolInfo
+    })
+}
+
+export function define<
+  Parameters extends Schema.Decoder<unknown>,
+  Result extends Metadata,
+  R,
+  ID extends string = string,
+>(
+  id: ID,
+  init: Effect.Effect<Init<Parameters, Result>, never, R>,
+): Effect.Effect<Info<Parameters, Result>, never, R | Truncate.Service | Agent.Service> & { id: ID } {
+  return Object.assign(
+    Effect.gen(function* () {
+      const resolved = yield* init
+      const truncate = yield* Truncate.Service
+      const agents = yield* Agent.Service
+      return { id, init: wrap(id, resolved, truncate, agents) }
+    }),
+    { id },
+  )
+}
+
+export function init<P extends Schema.Decoder<unknown>, M extends Metadata>(
+  info: Info<P, M>,
+): Effect.Effect<Def<P, M>> {
+  return Effect.gen(function* () {
+    const init = yield* info.init()
+    return {
+      ...init,
+      id: info.id,
+    }
+  })
+}
+
+export * as Tool from "./tool"

package/src/tool/truncate.ts

@@ -0,0 +1,156 @@
+import { LayerNode } from "@msrfteam/core/effect/layer-node"
+import { NodePath } from "@effect/platform-node"
+import { Cause, Duration, Effect, Layer, Option, Schedule, Context } from "effect"
+import path from "path"
+import type { Agent } from "../agent/agent"
+import { FSUtil } from "@msrfteam/core/fs-util"
+import { evaluate } from "@/permission/evaluate"
+import { Config } from "@/config/config"
+import { ToolID } from "./schema"
+import { TRUNCATION_DIR } from "./truncation-dir"
+
+const RETENTION = Duration.days(7)
+
+export const MAX_LINES = 2000
+export const MAX_BYTES = 50 * 1024
+export const DIR = TRUNCATION_DIR
+export const GLOB = path.join(TRUNCATION_DIR, "*")
+
+export type Result = { content: string; truncated: false } | { content: string; truncated: true; outputPath: string }
+
+export interface Options {
+  maxLines?: number
+  maxBytes?: number
+  direction?: "head" | "tail"
+}
+
+function hasTaskTool(agent?: Agent.Info) {
+  if (!agent?.permission) return false
+  return evaluate("task", "*", agent.permission).action !== "deny"
+}
+
+export interface Interface {
+  readonly cleanup: () => Effect.Effect<void>
+  readonly write: (text: string) => Effect.Effect<string>
+  /**
+   * Returns output unchanged when it fits within the limits, otherwise writes the full text
+   * to the truncation directory and returns a preview plus a hint to inspect the saved file.
+   */
+  readonly output: (text: string, options?: Options, agent?: Agent.Info) => Effect.Effect<Result>
+  /**
+   * Resolved truncation limits: values from `tool_output` in mervex config, or MAX_LINES / MAX_BYTES if unset.
+   */
+  readonly limits: () => Effect.Effect<{ maxLines: number; maxBytes: number }>
+}
+
+export class Service extends Context.Service<Service, Interface>()("@mervex/Truncate") {}
+
+const layer = Layer.effect(
+  Service,
+  Effect.gen(function* () {
+    const fs = yield* FSUtil.Service
+
+    const cleanup = Effect.fn("Truncate.cleanup")(function* () {
+      const cutoff = Date.now() - Duration.toMillis(RETENTION)
+      const entries = yield* fs.readDirectory(TRUNCATION_DIR).pipe(
+        Effect.map((all) => all.filter((name) => name.startsWith("tool_"))),
+        Effect.catch(() => Effect.succeed([])),
+      )
+      for (const entry of entries) {
+        const file = path.join(TRUNCATION_DIR, entry)
+        const info = yield* fs.stat(file).pipe(Effect.catch(() => Effect.succeed(undefined)))
+        const mtime = info && Option.getOrUndefined(info.mtime)
+        if (!mtime || mtime.getTime() >= cutoff) continue
+        yield* fs.remove(file).pipe(Effect.catch(() => Effect.void))
+      }
+    })
+
+    const write = Effect.fn("Truncate.write")(function* (text: string) {
+      const file = path.join(TRUNCATION_DIR, ToolID.ascending())
+      yield* fs.ensureDir(TRUNCATION_DIR).pipe(Effect.orDie)
+      yield* fs.writeFileString(file, text).pipe(Effect.orDie)
+      return file
+    })
+
+    const limits = Effect.fn("Truncate.limits")(function* () {
+      const configSvc = yield* Effect.serviceOption(Config.Service)
+      if (Option.isNone(configSvc)) return { maxLines: MAX_LINES, maxBytes: MAX_BYTES }
+      const cfg = yield* configSvc.value.get().pipe(Effect.catch(() => Effect.succeed(undefined)))
+      return {
+        maxLines: cfg?.tool_output?.max_lines ?? MAX_LINES,
+        maxBytes: cfg?.tool_output?.max_bytes ?? MAX_BYTES,
+      }
+    })
+
+    const output = Effect.fn("Truncate.output")(function* (text: string, options: Options = {}, agent?: Agent.Info) {
+      const resolved = yield* limits()
+      const maxLines = options.maxLines ?? resolved.maxLines
+      const maxBytes = options.maxBytes ?? resolved.maxBytes
+      const direction = options.direction ?? "head"
+      const lines = text.split("\n")
+      const totalBytes = Buffer.byteLength(text, "utf-8")
+
+      if (lines.length <= maxLines && totalBytes <= maxBytes) {
+        return { content: text, truncated: false } as const
+      }
+
+      const out: string[] = []
+      let i = 0
+      let bytes = 0
+      let hitBytes = false
+
+      if (direction === "head") {
+        for (i = 0; i < lines.length && i < maxLines; i++) {
+          const size = Buffer.byteLength(lines[i], "utf-8") + (i > 0 ? 1 : 0)
+          if (bytes + size > maxBytes) {
+            hitBytes = true
+            break
+          }
+          out.push(lines[i])
+          bytes += size
+        }
+      } else {
+        for (i = lines.length - 1; i >= 0 && out.length < maxLines; i--) {
+          const size = Buffer.byteLength(lines[i], "utf-8") + (out.length > 0 ? 1 : 0)
+          if (bytes + size > maxBytes) {
+            hitBytes = true
+            break
+          }
+          out.unshift(lines[i])
+          bytes += size
+        }
+      }
+
+      const removed = hitBytes ? totalBytes - bytes : lines.length - out.length
+      const unit = hitBytes ? "bytes" : "lines"
+      const preview = out.join("\n")
+      const file = yield* write(text)
+
+      const hint = hasTaskTool(agent)
+        ? `The tool call succeeded but the output was truncated. Full output saved to: ${file}\nUse the Task tool to have explore agent process this file with Grep and Read (with offset/limit). Do NOT read the full file yourself - delegate to save context.`
+        : `The tool call succeeded but the output was truncated. Full output saved to: ${file}\nUse Grep to search the full content or Read with offset/limit to view specific sections.`
+
+      return {
+        content:
+          direction === "head"
+            ? `${preview}\n\n...${removed} ${unit} truncated...\n\n${hint}`
+            : `...${removed} ${unit} truncated...\n\n${hint}\n\n${preview}`,
+        truncated: true,
+        outputPath: file,
+      } as const
+    })
+
+    yield* cleanup().pipe(
+      Effect.catchCause((cause) => Effect.logError("truncation cleanup failed", { cause: Cause.pretty(cause) })),
+      Effect.repeat(Schedule.spaced(Duration.hours(1))),
+      Effect.delay(Duration.minutes(1)),
+      Effect.forkScoped,
+    )
+
+    return Service.of({ cleanup, write, output, limits })
+  }),
+)
+
+export const node = LayerNode.make({ service: Service, layer: layer, deps: [FSUtil.node] })
+
+export * as Truncate from "./truncate"

package/src/tool/truncation-dir.ts

@@ -0,0 +1,4 @@
+import path from "path"
+import { Global } from "@msrfteam/core/global"
+
+export const TRUNCATION_DIR = path.join(Global.Path.data, "tool-output")

package/src/tool/video-gen.ts

@@ -0,0 +1,50 @@
+import * as Tool from "./tool"
+import { Effect, Schema } from "effect"
+import { Config } from "@/config/config"
+
+const Parameters = Schema.Struct({
+  prompt: Schema.String.annotate({ description: "Description of the video to generate" }),
+})
+
+async function callModel(apiKey: string, model: string, prompt: string, maxTokens = 2000): Promise<string> {
+  const res = await fetch("https://api.mervex.cc.cd/v1/chat/completions", {
+    method: "POST",
+    headers: { "Content-Type": "application/json", Authorization: `Bearer ${apiKey}` },
+    body: JSON.stringify({ model, stream: true, max_tokens: maxTokens, messages: [{ role: "user", content: prompt }] }),
+  })
+  if (!res.ok) {
+    const text = await res.text().catch(() => "")
+    throw new Error(`mervex API ${res.status}: ${text.slice(0, 300)}`)
+  }
+  const text = await res.text()
+  const lines = text.split("\n").filter((l) => l.startsWith("data: ") && l !== "data: [DONE]")
+  let out = ""
+  for (const line of lines) {
+    try {
+      const json = JSON.parse(line.slice(6))
+      const delta = json.choices?.[0]?.delta?.content
+      if (delta) out += delta
+    } catch {}
+  }
+  return out || text
+}
+
+export const VideoGenTool = Tool.define(
+  "video_gen",
+  Effect.gen(function* () {
+    const config = yield* Config.Service
+    return {
+      description:
+        "Generate a video from a text description using AI. Use when the user wants to create, generate, or produce a video.",
+      parameters: Parameters,
+      execute: (params: Schema.Schema.Type<typeof Parameters>, ctx: Tool.Context) =>
+        Effect.gen(function* () {
+          const cfg = yield* config.get()
+          const apiKey = cfg.provider?.mervex?.options?.apiKey
+          if (!apiKey) return yield* Effect.fail(new Error("No API key configured for mervex provider"))
+          const output = yield* Effect.promise(() => callModel(apiKey, "Msrfteam/Mervex-ai-v0.1-vd", params.prompt))
+          return { title: "Generated Video", output, metadata: {} }
+        }).pipe(Effect.orDie),
+    }
+  }),
+)

package/src/tool/webfetch.ts

@@ -0,0 +1,192 @@
+import { Effect, Schema } from "effect"
+import { HttpClient, HttpClientRequest } from "effect/unstable/http"
+import { Parser } from "htmlparser2"
+import * as Tool from "./tool"
+import TurndownService from "turndown"
+import DESCRIPTION from "./webfetch.txt"
+import { isImageAttachment } from "@/util/media"
+
+const MAX_RESPONSE_SIZE = 5 * 1024 * 1024 // 5MB
+const DEFAULT_TIMEOUT = 30 * 1000 // 30 seconds
+const MAX_TIMEOUT = 120 * 1000 // 2 minutes
+
+export const Parameters = Schema.Struct({
+  url: Schema.String.annotate({ description: "The URL to fetch content from" }),
+  format: Schema.Literals(["text", "markdown", "html"])
+    .annotate({
+      description: "The format to return the content in (text, markdown, or html). Defaults to markdown.",
+      default: "markdown",
+    })
+    .pipe(Schema.withDecodingDefault(Effect.succeed("markdown" as const))),
+  timeout: Schema.optional(Schema.Number).annotate({ description: "Optional timeout in seconds (max 120)" }),
+})
+
+export const WebFetchTool = Tool.define(
+  "webfetch",
+  Effect.gen(function* () {
+    const http = yield* HttpClient.HttpClient
+    const httpOk = HttpClient.filterStatusOk(http)
+
+    return {
+      description: DESCRIPTION,
+      parameters: Parameters,
+      execute: (params: Schema.Schema.Type<typeof Parameters>, ctx: Tool.Context) =>
+        Effect.gen(function* () {
+          if (!params.url.startsWith("http://") && !params.url.startsWith("https://")) {
+            throw new Error("URL must start with http:// or https://")
+          }
+
+          yield* ctx.ask({
+            permission: "webfetch",
+            patterns: [params.url],
+            always: ["*"],
+            metadata: {
+              url: params.url,
+              format: params.format,
+              timeout: params.timeout,
+            },
+          })
+
+          const timeout = Math.min((params.timeout ?? DEFAULT_TIMEOUT / 1000) * 1000, MAX_TIMEOUT)
+
+          // Build Accept header based on requested format with q parameters for fallbacks
+          let acceptHeader = "*/*"
+          switch (params.format) {
+            case "markdown":
+              acceptHeader = "text/markdown;q=1.0, text/x-markdown;q=0.9, text/plain;q=0.8, text/html;q=0.7, */*;q=0.1"
+              break
+            case "text":
+              acceptHeader = "text/plain;q=1.0, text/markdown;q=0.9, text/html;q=0.8, */*;q=0.1"
+              break
+            case "html":
+              acceptHeader =
+                "text/html;q=1.0, application/xhtml+xml;q=0.9, text/plain;q=0.8, text/markdown;q=0.7, */*;q=0.1"
+              break
+            default:
+              acceptHeader =
+                "text/html,application/xhtml+xml,application/xml;q=0.9,image/avif,image/webp,image/apng,*/*;q=0.8"
+          }
+          const headers = {
+            "User-Agent":
+              "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/143.0.0.0 Safari/537.36",
+            Accept: acceptHeader,
+            "Accept-Language": "en-US,en;q=0.9",
+          }
+
+          const request = HttpClientRequest.get(params.url).pipe(HttpClientRequest.setHeaders(headers))
+
+          // Retry with honest UA if blocked by Cloudflare bot detection (TLS fingerprint mismatch)
+          const response = yield* httpOk.execute(request).pipe(
+            Effect.catchIf(
+              (err) =>
+                err.reason._tag === "StatusCodeError" &&
+                err.reason.response.status === 403 &&
+                err.reason.response.headers["cf-mitigated"] === "challenge",
+              () =>
+                httpOk.execute(
+                  HttpClientRequest.get(params.url).pipe(
+                    HttpClientRequest.setHeaders({ ...headers, "User-Agent": "mervex" }),
+                  ),
+                ),
+            ),
+            Effect.timeoutOrElse({ duration: timeout, orElse: () => Effect.die(new Error("Request timed out")) }),
+          )
+
+          // Check content length
+          const contentLength = response.headers["content-length"]
+          if (contentLength && parseInt(contentLength) > MAX_RESPONSE_SIZE) {
+            throw new Error("Response too large (exceeds 5MB limit)")
+          }
+
+          const arrayBuffer = yield* response.arrayBuffer
+          if (arrayBuffer.byteLength > MAX_RESPONSE_SIZE) {
+            throw new Error("Response too large (exceeds 5MB limit)")
+          }
+
+          const contentType = response.headers["content-type"] || ""
+          const mime = contentType.split(";")[0]?.trim().toLowerCase() || ""
+          const title = `${params.url} (${contentType})`
+
+          if (isImageAttachment(mime)) {
+            const base64Content = Buffer.from(arrayBuffer).toString("base64")
+            return {
+              title,
+              output: "Image fetched successfully",
+              metadata: {},
+              attachments: [
+                {
+                  type: "file" as const,
+                  mime,
+                  url: `data:${mime};base64,${base64Content}`,
+                },
+              ],
+            }
+          }
+
+          const content = new TextDecoder().decode(arrayBuffer)
+
+          // Handle content based on requested format and actual content type
+          switch (params.format) {
+            case "markdown":
+              if (contentType.includes("text/html")) {
+                const markdown = convertHTMLToMarkdown(content)
+                return {
+                  output: markdown,
+                  title,
+                  metadata: {},
+                }
+              }
+              return { output: content, title, metadata: {} }
+
+            case "text":
+              if (contentType.includes("text/html")) {
+                return { output: extractTextFromHTML(content), title, metadata: {} }
+              }
+              return { output: content, title, metadata: {} }
+
+            case "html":
+              return { output: content, title, metadata: {} }
+
+            default:
+              return { output: content, title, metadata: {} }
+          }
+        }).pipe(Effect.orDie),
+    }
+  }),
+)
+
+function extractTextFromHTML(html: string) {
+  let text = ""
+  let skipDepth = 0
+
+  const parser = new Parser({
+    onopentag(name) {
+      if (skipDepth > 0 || ["script", "style", "noscript", "iframe", "object", "embed"].includes(name)) {
+        skipDepth++
+      }
+    },
+    ontext(input) {
+      if (skipDepth === 0) text += input
+    },
+    onclosetag() {
+      if (skipDepth > 0) skipDepth--
+    },
+  })
+
+  parser.write(html)
+  parser.end()
+
+  return text.trim()
+}
+
+function convertHTMLToMarkdown(html: string): string {
+  const turndownService = new TurndownService({
+    headingStyle: "atx",
+    hr: "---",
+    bulletListMarker: "-",
+    codeBlockStyle: "fenced",
+    emDelimiter: "*",
+  })
+  turndownService.remove(["script", "style", "meta", "link"])
+  return turndownService.turndown(html)
+}

package/src/tool/webfetch.txt

@@ -0,0 +1,13 @@
+- Fetches content from a specified URL
+- Takes a URL and optional format as input
+- Fetches the URL content, converts to requested format (markdown by default)
+- Returns the content in the specified format
+- Use this tool when you need to retrieve and analyze web content
+
+Usage notes:
+  - IMPORTANT: if another tool is present that offers better web fetching capabilities, is more targeted to the task, or has fewer restrictions, prefer using that tool instead of this one.
+  - The URL must be a fully-formed valid URL
+  - HTTP URLs will be automatically upgraded to HTTPS
+  - Format options: "markdown" (default), "text", or "html"
+  - This tool is read-only and does not modify any files
+  - Results may be summarized if the content is very large