What it does
Its System One path uses a finite action menu; other agents keep their own loops. Session workspaces share one container, and self-hosting still relies on remote model providers.
Agent tooling
HarnessRouter
Run different agent runtimes through one task interface, including a Jev-driven finite-action controller.
Bookmark: HarnessRouter
Keep this in your collection.
Leave a noteWhat would you try with this?
: HarnessRouter
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
How you can use it
Have a developer open the Console, the tool’s control panel, and choose an AI assistant and a task. Follow its progress and open the files it produces in the same session. You can try another assistant through the same interface, though each works in its own way. Running the server yourself still involves requests to outside AI services.