@git-fabric/chat

Instructions

Implementation Reference

• src/layers/search.ts:13-33 (handler)

  The core search function that implements the chat_search tool logic. It takes a ChatAdapter, query string, and options (project, sessionId, limit), embeds the query using the adapter, and performs vector search on stored messages.

  export async function search(
    adapter: ChatAdapter,
    query: string,
    opts: {
      project?: string;
      sessionId?: string;
      limit?: number;
    },
  ): Promise<SearchResult[]> {
    const limit = opts.limit ?? 10;
  
    // Embed the query text
    const queryVector = await adapter.embed(query);
  
    // Search Qdrant
    return adapter.searchMessages(queryVector, {
      project: opts.project,
      sessionId: opts.sessionId,
      limit,
    });
  }

• src/app.ts:239-271 (registration)

  Registration of the chat_search tool in the tools array. Defines the tool name, description, input schema (with query, project, limit, sessionId parameters), and delegates execution to layers.search.search().

  {
    name: "chat_search",
    description:
      "Semantic search over all stored conversation content using vector similarity. Finds messages relevant to the query even if exact words don't match. Optionally scope to a project or specific session.",
    inputSchema: {
      type: "object",
      properties: {
        query: {
          type: "string",
          description: "Natural language search query.",
        },
        project: {
          type: "string",
          description: "Scope search to sessions with this project tag.",
        },
        limit: {
          type: "number",
          description: "Maximum number of results. Default: 10.",
        },
        sessionId: {
          type: "string",
          description: "Scope search to a single session UUID.",
        },
      },
      required: ["query"],
    },
    execute: async (args) =>
      layers.search.search(adapter, args.query as string, {
        project: args.project as string | undefined,
        sessionId: args.sessionId as string | undefined,
        limit: args.limit as number | undefined,
      }),
  },

• src/app.ts:243-264 (schema)

  Input schema definition for the chat_search tool. Validates required 'query' parameter and optional 'project', 'limit', and 'sessionId' parameters with their types and descriptions.

  inputSchema: {
    type: "object",
    properties: {
      query: {
        type: "string",
        description: "Natural language search query.",
      },
      project: {
        type: "string",
        description: "Scope search to sessions with this project tag.",
      },
      limit: {
        type: "number",
        description: "Maximum number of results. Default: 10.",
      },
      sessionId: {
        type: "string",
        description: "Scope search to a single session UUID.",
      },
    },
    required: ["query"],
  },

• src/types.ts:58-60 (schema)

  Type definition for SearchResult output type, which extends ChatMessage with a score field for semantic search relevance ranking.

  export interface SearchResult extends ChatMessage {
    score: number;
  }