Straight answers. No marketing language.
Mnemo runs entirely on your machine. Mem0, Letta, and Zep are cloud services that store your memory on their servers. Mnemo stores everything locally in ChromaDB and embeds your memories with a local model — no API keys, no accounts. Your data never leaves your laptop.
Yes. The memory database, embeddings, and search all run on your hardware — the core system makes no external calls and needs no API keys. The research and briefing automation added in the Done-With-You install uses only the web sources you choose to connect.
You have an export. Run mnemo export to get a timestamped zip of your entire memory database, then mnemo import backup.zip on the new machine — and you're back in business. No vendor lock-in.
Yes. Mnemo is built as a companion to Claude Desktop. It injects your local memory context into Claude sessions via the MCP protocol. Without Claude Desktop, you won't have the conversational interface that reads your memory.
Yes. mnemo export produces a portable zip you can move to any machine. You own the ChromaDB files, the code, and the configuration. Nothing is stored in a proprietary format.
$199 is a self-serve kit: the memory engine, CLI, MCP server, installer, smoke test, and documentation. You set it up yourself. $997 is a guided install: a 90-minute call where I configure everything on your machine, tune your research goals, and walk you through the system. Capped at four installs per month.
14 days from purchase. If the system doesn't work on your machine or you change your mind, email within 14 days and I'll refund. After 14 days, no refunds — you own the files and can continue using them.
No. Mnemo is a one-time purchase. There is no recurring fee, no cloud subscription, and no usage-based billing. You pay once, you own the system.
Yes. The Memory Concierge waitlist is for ongoing maintenance and custom research pipelines. If you need something outside the standard install, join the waitlist and we'll discuss scope.
ChromaDB on a laptop can comfortably hold tens of thousands of memory entries. The practical limit is your disk space and how much context you want Claude to read at the start of each session. Most users stay well under 100 MB of vector data.
No. The core system sends nothing — telemetry is disabled and no memory content ever leaves your machine. One exception worth knowing: the first run downloads the local embedding model once (~80 MB, from a public model repository). After that, everything works fully offline.
The README and smoke test are designed to surface problems early. If you're still blocked, the Done-With-You tier includes a follow-up check-in. Starter purchasers can email with specific errors; I respond to genuine setup questions.