Install in minutes

Download the MnemosyneC installer from mnemosynec.ai and run it (~456 MB). On first launch, MnemosyneC sets up a local AI engine — a free, open-source model running entirely via Ollama (MIT-licensed). One-time setup. No API keys. No cloud account. No subscription required to get started.

Want the best local model? Upgrade to Google Gemma 4 12B inside the app — free, MIT-licensed, ~7 GB one-time download, pulled via Ollama.

Your first session

1. Cover screen — select “Just use it” for immediate access, or walk through the full setup to configure your workspace and cooperative membership.
2. Ask anything — MnemosyneC runs entirely on your hardware. Your questions never leave your machine. No internet required after installation.
3. Add your files — optionally point MnemosyneC to a folder you want to remember. Read-only: your original files are never modified or uploaded. Ever.
4. Connect to the mesh — optional but powerful. Your node joins a cooperative knowledge network. 20 nodes tested at 16.6 ms p50 (BP067, hash-verified). Only what you choose to share, shares.

Local AI: how it actually works

MnemosyneC uses Ollama (MIT-licensed, open-source, by Jeffrey Morgan and Michael Chiang / Y Combinator) to run AI models directly on your CPU or GPU. The model weights live on your disk. Inference runs locally. Zero tokens are sent to any external server.

The design philosophy is “interest-knowledge, not intelligence” — MnemosyneC is the substrate that remembers your context, your notes, and your workspace structure. The local AI provides the reasoning layer. Together, you get a personalized AI assistant that learns your world without ever handing that world to a cloud provider.

The benchmark proof: a free, local Ollama model goes from 6% to 78% accuracy with the MnemosyneC substrate applied — a +72 percentage point gain (Cohen’s κ 0.936, BP065, hash-verified). That’s not marketing. You can run the test yourself.

Cooperative class: built different

MnemosyneC is built on the Liana Banyan cooperative platform. These aren’t marketing promises — they’re structural commitments locked into the corporate bylaws.

No advertising. No venture capital. No surveillance. Cooperative-AI is structural defense against hostile capture. When you use MnemosyneC, you’re not a user — you’re a member of a system designed to serve you, not harvest you.

License: free forever

MnemosyneC ships under SSPL v1 (Server Side Public License) + Cooperative Defensive Patent Pledge #2260. That means:

• Run it. No account, no subscription, no permission needed.
• Modify it. Source is auditable. Fork it for your community.
• Self-host it. Any cooperative can run their own.
• The pledge is irrevocable. No future board, acquirer, or successor can weaponize these patents against cooperative-class users.

21 USPTO provisional patent filings protect the commons — not to fence it, but to ensure no one else can.