mohamedhussien-zseeker/contextledger

ContextLedger describes itself as Git for AI memory. It gives AI agents and assistants a persistent, local place to store project knowledge that survives across chat sessions, so information does not get lost every time a conversation compacts or restarts. The memory is stored as plain Markdown files you can read and edit yourself, and indexed into a local SQLite database for fast searching. The project targets a specific problem: AI coding sessions forget the state of a project once older messages get summarized or dropped, documentation drifts away from what is actually true, and AI agents often retrieve information without being able to explain where it came from. ContextLedger's answer is a loop of capturing information, processing it into linked notes, indexing it, and making it searchable and servable to agents. Nothing needs to be sent to the cloud since everything runs locally on your own machine. You can interact with your stored memory in several ways: a command line tool for people and scripts, a REST API for other applications, an MCP server so AI agents that support the Model Context Protocol can query it directly, or by editing the underlying Markdown files by hand, including inside note-taking apps like Obsidian. There is also a small clipboard tool for pasting retrieved context into mobile AI chat apps. Search uses a relevance-scored full text search built into SQLite, and the system tracks links between notes to build a small knowledge graph with health metrics showing how connected the notes are. Getting started involves cloning the repository and running an install script, then using a handful of commands to create a memory vault, capture a source file, process it into notes, index it, and search it. It requires Python 3.11 or later, and the project is released under the GPL-3.0 license, with 93 automated tests reported passing.