TokenKeeper FAQs

Question 1

Is TokenKeeper processing done locally or in the cloud?

Accepted Answer

TokenKeeper is designed for local-first processing. It uses Ollama for embeddings and ChromaDB for vector storage directly on your machine. No cloud services or API keys are required, ensuring your data remains private and secure.

Question 2

What kind of search capabilities does TokenKeeper offer?

Accepted Answer

TokenKeeper features a powerful hybrid search, combining semantic similarity (vector embeddings) with keyword matching (BM25). Results are merged using Reciprocal Rank Fusion, providing both conceptual understanding and precise term recall.

Question 3

What is TokenKeeper?

Accepted Answer

TokenKeeper is an MCP server that provides local RAG (Retrieval-Augmented Generation) memory for AI coding agents. It indexes your project's code and documents to allow agents to query only relevant information, drastically reducing prompt tokens.

Question 4

How does TokenKeeper reduce AI agent token costs?

Accepted Answer

Instead of loading entire project files into an AI agent's context, TokenKeeper enables agents to query for only relevant chunks of information. This process can reduce prompt tokens by up to 80%, keeping agents in the high-quality zone of their context window.

Question 5

What types of files can TokenKeeper index?

Accepted Answer

TokenKeeper supports indexing a wide range of files including markdown (.md, .mdx), JSON, and source code files (e.g., Python, TypeScript, JavaScript, Go, Rust, Java, Ruby, C/C++). It uses heading-aware chunking for docs and AST parsing for code.

TokenKeeper

TokenKeeper

主要功能

使用案例

主要功能

使用案例