Memoir Ai

Memoir is memory infrastructure for AI agents. It presents itself as Git for AI memory, with a local-first, taxonomy-structured, Git-versioned store that lets agents explain, rewind, and branch their memory. The product is framed around problems such as context contamination, token rent, and memory drift.

Its core capabilities include recall by path rather than similarity, time travel to reproduce bugs, and branching to test risky strategies. The page also describes automated branch shadowing, where memory branches follow git branches automatically via Claude Code hooks, and a merge workflow that moves lessons from a feature branch into the main knowledge base. Memoir also supports semantic path retrieval with hierarchical paths such as api.v2.auth. The text says it ships as a Claude Code plugin with automatic skills and hooks that follow git workflow.

Memoir is built for coding agents and custom runtimes. It is shown with plugins or integrations for Claude Code, Codex, OpenCode, Hermes, OpenClaw, and LangGraph, and it also offers a Python SDK, a CLI, and an MCP Server for any MCP host. The installation examples include pip install memoir-ai for Python, and the page says the Python install is for the SDK or CLI. It also describes slash commands and hooks for session memory capture, context injection on start, and explicit recall or onboarding actions.

The page lists Apache 2.0 licensing and Python 3.10+ support. It also refers to open-source coding agent support in the plugin ecosystem and to community-maintained plugins for some runtimes.