Behavioral safety classifier for human-AI conversations. Eight user-risk axes, four AI-concern axes, fiction strength, per-turn trajectory. Self-hosted Docker, single GPU.
Free under Apache License 2.0. No license key. No startup gate. Run it however you want, modify it, redistribute it.
For production engagement (calibration / clinical / incident response / indemnification / ongoing model updates): [email protected].
Fill out the form and you'll see download links to the platform tarball + customer bundle. The form is for our records (so we know who's using it); the download itself is free.
Current OSS release: 04e6dee