Memmy Agent

Every coding agent starts from zero. Memmy is the layer that stops that happening: it reads the local history your AI tools have already written — Claude Code, Cursor, Codex, OpenClaw, Hermes Agent — turns it into structured, searchable memory, and injects the relevant slice back in when you move between tools. The pitch is literally "let every AI remember the same you": a decision you explained to Cursor last week is available to Claude Code today without you retelling it.

The flow is three steps. Scan imports history from tools you explicitly authorise. Organise runs the material through MemOS, the project's structured memory engine, which merges duplicates and builds retrievable memories rather than a flat transcript archive. Inject hands the right context to whichever agent is asking. A first-encounter report shows you what it learned about you before anything is shared.

Memory stays on the machine — nothing is uploaded — and the permission model is per-source and per-agent, so you can see which agents can read which memories, and view, disable or delete any source. Beyond the desktop app there is a memmy CLI with an OpenAI-compatible API server, and a standalone memmy-memory CLI so external agents and scripts can query the same store. Tool integrations cover Telegram, Discord, GitHub, Notion, MCP and custom skills.

Platforms: macOS 12+ and Windows 10+. Licence: MIT, open source, free — the desktop app runs either in account mode with trial credits or with your own API key.