we open-sourced mex today. it's the memory layer some of the newest yc companies are already running on, and it lives in git without a server.
the no-server part is the point. if your team's memory lives in the repo, it gets versioned with the code, branches with the code, and shows up in review next to the change that made it stale. the thing it describes and the description move together.
put memory in a separate service and it slowly turns into a second source of truth that nobody reviews. that's how you end up with an agent confidently describing a codebase that stopped existing three refactors ago.