Open by design.
Independent by implementation.

Why OpenJev exists

Agent workflows often need narrow judgments rather than free-form paragraphs: selecting a route, identifying an evidence gap or scoring a risk. OpenJev makes those judgments typed, while leaving workflow policy in ordinary code.

Project boundaries

OpenJev is an independent, clean-room project. It is not Jev and is not affiliated with or endorsed by TypeSafe AI. It does not contain TypeSafe AI’s proprietary model, weights, sampler, training code or private implementation.

The public API is independently implemented and is not advertised as a drop-in replacement for TypeSafe’s hosted service. Version 0.2 is a runtime, not a newly trained foundation model.

License and attribution

The OpenJev core is MIT licensed. Third-party providers, datasets and model weights remain subject to their own licenses.

MIT license ↗ · Third-party notices ↗

Contribute or get help

Report reproducible bugs and propose improvements through the project repository. Include the provider and version, a minimal non-sensitive example and the observed behavior.

For security-related reports, follow the repository’s security policy.

Website content checked against project source on September 23, 2026.