Praxion Labs

Praxion is an AI control engineer built on Meta Muse (muse-spark-1.3, Meta Model API). Give it a machine it has never seen. Muse designs small experiments, and a trusted supervisor checks each one before it touches the hardware. From the measured responses, Muse writes the controller itself, as Python. My code never edits it. The controller is tested in a locked-down sandbox on hidden scenarios the model never sees. When the machine changes, a monitor notices. Muse runs new experiments, names what changed and writes a repair. The repair ships only if it beats the original on the same hidden tests, with zero safety violations. We tested it at home on a $50 heater board, the same kind of thermal loop that cools a server hall. We switched a fan on to change the physics. Muse diagnosed the extra heat loss and rewrote its controller. Then both controllers ran on the physical board under the same fan, and the repair had 58% less error. Over ten hardware runs, eight of them unattended overnight, the monitor caught every change (8/8) and the diagnosis was right 7 of 7. All 4 repairs the gate promoted beat the original on the board, by 34 to 75%, with zero safety interventions. The failures are kept in the results. A heater board is small, but the loop is the same one that runs in server halls, reactors and water mains.