Perform a semantic hybrid search on the indexed source code.

This tool combines vector similarity (cosine) and full-text keyword search (BM25) to find the most relevant code snippets (chunks) based on the user's query intent. Use this tool when looking for specific implementations, where functions or features are defined, or to locate code matching a certain concept or task.

Retrieve a structured hierarchical tree map of classes, methods, and functions in the workspace.

This tool provides a compact overview of the codebase's architecture and logical structure. Use this tool to understand how the project is organized, list classes and methods within folders, or find entrypoints without retrieving full file contents. This is extremely token-efficient.

Get diagnostic metadata and health status of the local vector index.

This tool returns information such as whether the index exists, total indexed chunks, indexed repositories, active embedding model, last indexing timestamp, git HEAD SHA, and whether the index is stale compared to the current workspace code.

Index (or re-index) source code into the local search index.

IMPORTANT: unless the user already specified what to index, call with dry_run=true first, show the candidate folders to the user and ASK whether to index everything or specific folders. Only then call again with the chosen 'paths' (or none for the full workspace).

Indexing is incremental by default: unchanged files are skipped on subsequent calls, making re-runs fast. Use full=true to force a complete rebuild ignoring cached file hashes.