TokenKeeper

Example Use Cases

• 1

  Long-term project memory without exhausting context windows.

• 2

  That require efficient semantic and keyword search over large codebases and documentation.

• 3

  Local-first, privacy-preserving indexing and retrieval of project knowledge.

Description

TokenKeeper indexes your project's markdown, JSON, and source code files into a local vector database and exposes hybrid semantic and keyword search tools via MCP. It reduces prompt token usage by up to 80% by allowing AI agents to query only relevant chunks instead of loading entire files. The server runs fully locally using Ollama for embeddings and ChromaDB for vector storage, ensuring privacy and no cloud dependencies.