teflon07/memkeeper

Overview

Local-first, deterministic memory store for AI agents. Uses SQLite-backed BM25 with optional local ONNX semantic search and cross-encoder rerank, plus graph projection and an MCP adapter. Runs without any required network or LLM.

Best for

Best for
Developers building local-first AI agents that need deterministic memory without cloud dependencies.

Use cases

• Give AI agents persistent local memory without external APIs
• Run hybrid retrieval combining BM25 and offline semantic reranking
• Expose agent memory to MCP-compatible clients

Notes

Local-first, deterministic memory store for AI agents. Uses SQLite-backed BM25 with optional local ONNX semantic search and cross-encoder rerank, plus graph projection and an MCP adapter. Runs without any required network or LLM.

2 stars on GitHub. Last updated 2026-08-24. Licensed Apache-2.0.

Use cases

• Give AI agents persistent local memory without external APIs
• Run hybrid retrieval combining BM25 and offline semantic reranking
• Expose agent memory to MCP-compatible clients

Pros

• Fully offline with no required network or LLM
• Deterministic behavior for reproducible agent memory
• Rust implementation for performance and safe concurrency

Cons

• Very early stage with only 2 GitHub stars
• Requires Rust toolchain to build or integrate
• ONNX semantic search needs local model setup

Indexed from awesome-mcp-servers-punkpeye and enriched against its public facts.