The conversation that doesn't forget
Memu is a bot with long-term memory across sessions, devices, and model providers. Backed by the ZynContext G4-LENS Digital Twin, it ingests every conversation, extracts the durable facts, and rehydrates them at the start of every new chat. Last week's project brief is still there. So is your name. So is the joke you told on Tuesday.
Twenty-one source vectors, one digital twin
Memu uses the ZynRip protocol to extract information from conversations into twenty-one distinct source vectors — facts, preferences, open commitments, people mentioned, unresolved questions, style markers, and more. These are stored in the ZynContext engine, synchronised four ways across VPS, GitHub, Drive, and Notion, and retrieved on every turn via the Ground Truth Protocol.
Founders, researchers, writers, people
Anyone who has ever had a multi-week project stall because the chatbot forgot the premise. Anyone who has had to paste the same three paragraphs of context into every new ChatGPT session. Anyone whose work is actually the long arc and not the one-shot reply. Memu is built for the people who treat their assistants like collaborators and were never being served by session-scoped memory.
Your memory, your machine, optional cloud
The full Memu memory store can run on your own machine — the ZynContext engine is a self-hostable binary. The hosted tier exists for convenience and syncs are end-to-end encrypted to your key. Nothing in your memory store is used to train a model, cross-account, or ever.
Request a seat
Memu is in closed beta during the Zynthio foundation phase. Seats are allocated manually. Email corey@coreyai.ai describing the long-running work you are trying to do and why session-scoped memory has failed you.