Skip to main content

Search Tools

Read-only search tools run in parallel when the LLM calls several in one turn.

glob​

Finds files matching a glob pattern in the workspace.

Annotations: read-only · idempotent

Parameters:

  • pattern (string, required) — glob pattern, e.g. **/*.ts, src/**/*.test.ts
  • path (string, optional) — root directory to search (default: workspace root)

Uses rg --files -g <pattern> when ripgrep is available, falls back to in-process recursive scan.

Searches file contents for a text pattern or regex.

Annotations: read-only · idempotent

Parameters:

  • pattern (string, required) — text pattern or regex to search for
  • path (string, required) — directory to search

index_workspace​

Builds a TF-IDF vector index of all workspace files. Must be called once before semantic_search can be used. The index lives in memory for the session.

Annotations: none — mutates the in-memory vector index (session state), so it is not read-only and runs sequentially. See Tool Annotations.

Parameters:

  • extensions (string[], optional) — file extensions to include, e.g. [".ts", ".md"]. Defaults to common code and text extensions.

Respects .gitignore patterns. Re-run after large file changes to keep the index current.

Searches the indexed workspace for files or code chunks semantically relevant to a natural-language query.

Annotations: read-only · idempotent

Parameters:

  • query (string, required) — natural language query, e.g. "where do we handle JSON-RPC messages"
  • topK (number, optional) — maximum number of results to return (default: 5)
  • minScore (number, optional) — minimum cosine similarity threshold 0–1 (default: 0.1)

Returns ranked chunks with file path and relevance score. Requires index_workspace to have been called first.