ai-memory
Shared long-term memory that lets coding agents hand work off to each other.
- Stars
- 14.9k
- Forks
- 1.1k
- Licence
- MIT
- Language
- Rust
- Last update
- 5 days ago
- Type
- Dev tool
What it is
ai-memory keeps notes about your coding projects so that AI coding agents can pick up where another left off. You run its server yourself, and agent tools such as Claude Code or Codex feed it session activity through hooks, which it turns into searchable markdown pages. The README says a memory stored this way can be handed to a different agent or machine without re-explaining the work.
- Quit one agent mid-task and continue the same work in another agent in the same directory
- Run agents through `ai-memory run` so the next agent receives the shared workstream and a bounded startup brief
- Search project memory with full-text, entity, and link lookups from the command line or an agent
- Keep one shared memory server so a team's agents can retrieve what other members' sessions recorded
Great for
- Developers who switch between several AI coding agents in the same project
- People who want agent memory stored as plain markdown files they can read and edit
- Small teams that want to share project knowledge from one self-hosted server
- Developers who work across more than one machine and want their project context to follow them
Not ideal for
- People who want a hosted service with no server to run, since the README describes self-hosting
- Users of agent tools not on the support matrix, since integration is listed per tool
- Anyone who needs the optional LLM-based features without configuring an LLM, since the default path makes zero LLM calls and the 'dream' pass is off by default
How to get it
- 1
See the README
You'll need: Rust 1.95+ (per the README badge, if building from source)
Clone the repository
git clone https://github.com/akitaonrails/ai-memory - 2
Run it
- Install a release package (RPM or Claude plugin zip) from the v3.2.10 release, or build from source
- Start the server as described in the README's setup docs
- Launch an agent through `ai-memory run <agent>` so its sessions are captured
- Start a new agent in the same directory and check that it receives the handoff brief
- 3
Get the most out of it
- Enable the management console with `--features management-ui` and `serve --transport http --enable-management-ui` to administer users and sessions
- Review docs/cross-project-profile.md to decide whether to enable the default-on profile for your setup
- Read docs/support-matrix.md to see per-agent caveats before relying on a given integration
- Keep memory retention defaults unless you read the compaction and dedup notes, since those rewrite content and are off by default
Downloads
Strengths
- Memory is stored as ordinary git-backed markdown files that can be grepped, edited, and rebuilt from source
- Works with zero LLM calls and no API key by default for capture, search, and handoffs
- Ships as a self-contained binary with an audit log and built-in multi-user auth, according to the README
Trade-offs
- Requires running a server to share memory across machines or people, and the README does not walk through that setup in detail here
- Several support-matrix entries are MCP-only or hooks-only, so features vary by agent
- The README is truncated, so details on configuration, limits, and security model are not fully verifiable from this text
Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.