Why agent-memory
- Tell one agent, both remember. A preference you give Codex is there the next time Claude
Code starts, and the other way round, on every machine that pulls.
- The right memories in each session. Everything global, plus what was learned in this folder
or repository, and nothing from your other projects.
- Memory in git. One Markdown file per memory in a git repository you own, with where it came
from recorded by the hooks. Every change is a commit you can review or revert; add a private
remote to sync machines, or keep it all on one. Here is an example.
Profiles keep separate sets of
memories in repositories of their own, and a switch reaches running sessions too.
- Nothing to run. Two hooks and git. No service, no account; search by meaning runs on your CPU.
- A page to keep it tidy.
mem ui searches by words and by meaning, edits with undo, and shows
what changed week by week.
Quick start
See it first, with nothing installed:
uvx --from "agent-memory-git[ui]" mem ui --demo
Then set it up (Python 3.11+, git and uv):
uv tool install "agent-memory-git[ui]"
mem init ~/repos/agent-memory-data # your data repository
mem install --dry-run # every change, as a diff
mem install
Codex asks you to trust new hooks once: open /hooks in Codex and enable the three agent-memory
hooks. The docs cover importing existing
memories, every file mem install changes, upgrading and uninstalling.
How it works
At session start a hook gives the agent an index of the memories that apply where it runs. The
agent saves what is worth keeping as Markdown files, following
PROTOCOL.md, and at the end of each turn a hook records where each
came from, commits and pushes. The docs explain
the file format, where memories load, the commands and the configuration.
Contributing
uv sync --all-extras && uv run pytest
The mem ui page is plain HTML, CSS and JavaScript with no build step, and
uv run scripts/site.py serve runs the website and its live demo locally. More in the
docs.
License
MIT
