Free Tool Runs AI Decisions Privately on Your Own Laptop

A new free, open-source tool called Ollaya just launched, and it does something quietly useful: it runs "decision models" on your own computer. A decision model isn't a chatbot that writes paragraphs. It's a small AI that reads a piece of text or data and answers a narrow question — like "is this push risky?" or "does this customer want an invoice or a refund?" — and hands back an answer plus a confidence score. In one demo, it flagged a dangerous command as "destructive: yes" with 90% confidence in 178 milliseconds.

Why should you care? Speed and privacy. Ollaya answers a five-question request in about 10 milliseconds on a decent graphics card, versus roughly 236 to 276 milliseconds for the paid cloud service it mimics. That's the difference between a system that feels instant and one that lags. More importantly, your data never leaves the building. Customer emails, support tickets and chat messages — often the most sensitive things a company holds — get scored right where they already sit. The server only listens on your own machine by default.

Switching is meant to be easy. Ollaya speaks the same language as TypeSafe's existing API, so the official Python toolkit works unchanged — developers basically just change one web address. It ships with a menu of downloadable models: "laya" is the fastest and handles over 100 languages, "decider" is the most accurate, "qwen3guard" screens text for unsafe content in 119 languages, and "von" can read documents up to 8,192 tokens long.

The catch is real. This is a tool for technical teams, not a download-and-click app. It needs a reasonably capable computer, and it only answers structured, multiple-choice-style questions — it won't write essays or hold a conversation. Also, the speed comparison was measured differently on each side, so treat it as a rough order-of-magnitude claim rather than a precise scoreboard.