Your named memory. Persistent memory for coding agents, stored as git notes in the
repositories you already have, served over MCP and a CLI with identical commands. Personal
memory stays on your machine; distributed memory travels with git push. Sibling of
ynh: ynh manages how an agent is guided, ynm what it remembers.
brew install eyelock/tap/ynm # one executable, needs only git
cd your-repo && ynm init # distributed memory for this repo, and the agent clients it already usesynm init creates the distributed memory mount (personal memory lives in ~/.ynm) and configures the
agent clients this repository already uses: Claude Code, OpenCode or Pi. Each gets the MCP server,
the memory guidance, and, where the client has them, hooks that load memory when a session starts
and steer "remember this" into ynm rather than the client's own note files. A client found only on
your machine, such as Copilot CLI, gets a one-line suggestion, and a ynh harness gets ynm with
ynm client install ynh. ynm client install <client> does the same for one client by hand.
Other ways to install (a direct download, the slim build on your own Node, from source, the Docker image) are in Install ynm.
The agent now has memory_recall, memory_remember and eight more tools; every one is also a
CLI command:
ynm remember --type procedural --level distributed --content "Run pnpm check before pushing"
ynm recall --text "before pushing"ynm serve is the MCP server (stdio), ynm serve --http the hosted service.
eyelock.github.io/ynm (source in docs/):
- Tutorials: ordered lessons that double as the acceptance tests.
- How-to guides: hosted stores, keys, releases.
- Explanation: the six memory types, and why ynm works as it does.
- Reference: commands, tools, configuration.
Contributing, building and running the evals: CONTRIBUTING.md.
MIT