Free. MIT. Forever.

Yes. The library is MIT-licensed at HEAD and every tagged release. There is no feature gate. When the Enterprise tier ships, it will sell access to humans (named engineers, co-development), not access to features the open-source code does not already have.

No. The library runs in your process; nothing leaves your network unless you explicitly call out to a remote validator. We have no telemetry. No prompts, no completions, no PII, no IP addresses. Audit it yourself in `src/`.

Yes. The runtime is a pure Node module with no required network calls. Airgapped deployment is supported out of the box on the Open tier — no Enterprise-only path required.

No firm date. We are scoping the offer with design partners on GitHub Discussions before announcing a release timeline. If you have a production deployment that would benefit from a named engineer + co-development, please tell us what you need.

In-process, no network hop. The full engine (2 validators + 1 guard, short prompt) measures ~0.05 ms p50 on a single-core M-series — see packages/core/benchmarks/RESULTS.md for the methodology. Longer prompts scale linearly: medium prompt ~0.3 ms, ~5.6 KB long-form prompt ~14 ms. LLM-in-loop validators inherit upstream model latency separately. The pipeline runs in parallel where the contract allows.

Public advisories on GitHub at the time of release. Embargoed pre-disclosure and a private fix channel are on the Enterprise roadmap; until that ships, every reporter and consumer is on the same public timeline.