You spend an hour with Claude figuring out how your auth flow works, why that API returns 429s under load, or how to properly configure your test database. Real progress.
Next week, different session — you're explaining it all over again.
I built memo — a local memory layer for AI coding agents (Claude Code, Cursor, Codex, or anything MCP-compatible). It gives your AI assistant persistent, searchable memory across all sessions and projects.
Written in Go, runs 100% locally. Nothing leaves your machine!
Uses SQLite + vector search and an on-device embedding model, so it finds relevant memories even when you phrase things differently.
How I use it:
When I figure out why something works a certain way — an undocumented API quirk, a tricky config, a non-obvious design decision — I tell Claude to save it. Next time that context is needed, it's already there.
When onboarding onto a new codebase, I save architecture notes and gotchas as I learn them. Weeks later, Claude already has the context instead of me re-reading the same code.
Over time it becomes a personal knowledge base that your AI assistant actually uses — and it starts surfacing relevant context without you asking.
brew install ybonda/tap/memo
claude mcp add --scope user memo -- memo serve
That's it. Persistent memory across all your projects — everything saved locally in ~/.memo/memories.db
No comments:
Post a Comment