A shared hippocampus for your scattered AI agents

What it does Memmy runs locally as a desktop app, CLI, or API and acts as a persistent memory layer for AI agents. It ingests chat histories from tools like Cursor, Claude Code, and Codex, distills them into structured long-term memory, and serves that context back to any connected agent so you stop repeating yourself. It is essentially a personal knowledge base that follows you across assistants.

The interesting bit Rather than building another chatbot, Memmy treats memory as the foundation and agents as interchangeable consumers. Its MemOS engine turns scattered conversations into searchable context—though the README is light on the technical specifics of that distillation process.

Key highlights

• Cross-agent memory sharing: Cursor, Claude Code, Codex, OpenClaw, and Hermes Agent can all draw from the same context pool.
• Local-first by default: memory and configuration stay on your machine, with an explicit “no fake memories” policy when the store is unreachable.
• History onboarding: automatically imports past conversations from supported agents and generates a personalized “First Meeting Report.”
• Extensible runtime: supports MCP and custom Skills, plus managed Chromium browser tools and integrations for Telegram, Discord, Slack, GitHub, and more.
• BYOK-friendly: after free trial credits expire, switch to your own API keys with OpenAI-compatible model routing.

Caveats

• The README admits its feature comparison is based on public positioning, not item-by-item benchmarking.
• The documentation emphasizes local-first storage, but leaves unclear how “Account mode” and its cloud-hosted trial tokens interact with your data.

Verdict Developers juggling multiple AI coding tools and tired of re-explaining project context should look here; those already happy inside a single-agent ecosystem probably don’t need another runtime.