Close to home
The model runs on your machine and comes with the suite. Where information goes should be a deliberate choice.
An independent AI lab
Useful AI. On your own machine.
Built to act with you.
Models make possibilities.
We build what happens next.
AI should help you move through real work—with a clear view of what it knows, what it’s doing, and where your decision matters.
The model runs on your machine and comes with the suite. Where information goes should be a deliberate choice.
Meaningful actions deserve a moment of attention. Readable approvals keep the decision with the operator.
Page references, action outcomes, and retained evidence make the work easier to inspect and understand.
Inside Core
Follow a request through the system.
An illustrative interaction, not a live model.
The Core family
Every surface is named for what it may touch. The one that reads the web holds none of your files. The one that holds your documents has no web. Nothing is handed between them.
It proposes.
You decide.
A local assistant for the documents you hand it. It reads what you attach and can propose an action; the action waits for your approval, and what happened is kept on record. It has no web access.
One page in.
Its own words out.
Ask about a page you chose. Each sentence of the answer shows the words it quotes from that page, or says plainly that it is unsourced. It holds no files and takes no actions.
Rewriting and review inside a document, with the same approvals. Built as an experiment and not offered.
Documents and scans into tables, each value tied to the page it came from. Earlier work under the name NorthParse, now paused.
One install, with its model included. Your one choice: CPU or GPU.
An installer for Windows, then macOS, is being built. There is nothing to download yet.
Still building. Always curious.
Harness Core is a working lab. We’re turning experiments into a suite you can install, learning from real use, and making the machinery easier to live with.
Explore the system ↑