The principles behind Local.
Every closed model is a meter on your mind. It logs what you ask, learns what you fear, and sells the shape of your attention back to you as "safety." We think that's backwards.
Your prompt enters a sealed lane, runs on open weights, and the answer comes back. We keep no logs. We store no history. There is no account to subpoena, no profile to leak, no exhaust to sell. Close the tab and there is nothing left — not even to us.
We run open-weight models — DeepSeek, GLM, Gemma, Kimi and more — the ones you can inspect, not the ones that inspect you. Tuned to excel, not to flatter. The model works for you, not on you.
No engagement loops. No retention games. No dark patterns dressed as features. Local is built to answer the question in front of it and then forget you ever asked.
The same private lane serves a person at a keyboard and an autonomous agent over MCP. Your agents stop leaking every prompt to closed APIs that log them. Private inference, callable, settled on Base.
nothing leaves. — local