Local Git-versioned memory for AI coding agents

Capture, compile, recall via a local LLM wiki with on-device embeddings and an MCP server, no RAG or Docker.

Local Git-Versioned Memory for AI Coding Agents

This tool provides a robust, local memory solution for AI coding agents, leveraging Git for version control and an on-device LLM for recall. It eliminates the need for external RAG systems or Docker containers, simplifying deployment and management for developers. By storing agent memories as a local, Git-versioned wiki, it ensures data integrity, traceability, and efficient retrieval of past interactions and learned information.

What it Does

The core functionality is to create a persistent, version-controlled knowledge base for AI coding agents. It captures agent interactions, code snippets, and learned concepts, storing them in a structured, local format. This data can then be compiled and recalled using an on-device LLM, allowing the agent to access its history and context without relying on cloud services. The system utilizes on-device embeddings for efficient semantic search within the memory.

Key Features

Local Git Versioning: All memory data is managed under Git, providing a complete history of agent knowledge and enabling rollbacks.
On-Device LLM Integration: Facilitates local recall and compilation of memories without external API calls.
On-Device Embeddings: Enables fast and efficient semantic searching of stored information.
MCP Server Compatibility: Designed to integrate seamlessly with MCP servers, streamlining agent workflows.
No RAG or Docker Required: Simplifies setup and reduces dependencies, making it ideal for local development environments.

Who it's For

This tool is specifically designed for AI developers and researchers building and deploying coding agents. It is ideal for those who prioritize data privacy, require offline capabilities, or wish to maintain complete control over their agent's knowledge base. Developers working on complex projects requiring long-term memory and context for their AI agents will find this solution particularly beneficial. It's also suited for individuals looking to experiment with AI agent memory without the overhead of complex infrastructure.