ai-memory
Self-hosted memory for AI coding agents, with editable Markdown notes and cross-agent handoffs. Runs offline without an LLM or API key by default.
ai-memory keeps project knowledge on a server you control so developers can switch coding assistants or machines without losing context. It connects Claude Code, Codex, Cursor, Gemini CLI and OpenCode to shared memory, including unfinished work, failed approaches and open questions.
The memory lives in a git-backed Markdown wiki. You can read and edit the files directly, including in Obsidian, and rebuild the search database from them. Automatic capture records prompts, tool calls and session boundaries, sanitizes them before storage, and turns session activity into readable pages. Subsequent sessions receive a brief and can search by text, entities and links; vector search is optional.
It's open source under MIT. The Rust server runs on Linux, macOS and Windows through WSL2, with Docker and Podman options. Memory capture, search and handoffs work offline without an LLM or API key by default. Notes stay on your laptop or self-hosted server; optional LLM processing and embeddings can use external providers such as Anthropic and OpenAI.
Teams can share knowledge per project while keeping personal handoffs separate. Multi-user authentication, individual attribution and an audit log are built in. Repository rules can exclude sensitive paths from capture. Optional memory maintenance can compact older notes, combine near-duplicates and flag possible contradictions without LLM calls, while retaining original versions in git. A read-only browser view lets people inspect the wiki outside their coding assistant.
