Aurora-Lens gives your AI rules.
The model can change.
The rules don't.

Most AI governance lives inside the model. Aurora-Lens doesn't. It sits between your application and the model, blocking unsupported decisions before generation occurs.

Unsupported recommendation The AI recommends a course of action when the underlying evidence is unresolved.

The model has enough context to sound confident. The evidence does not support the conclusion.

Aurora-Lens refuses before the LLM is called. The declared uncertainty binds the gate.

Ambiguous referent The AI resolves a pronoun to a specific entity without grounds for which entity was meant.

Two entities are in scope. The model picks the more probable one and proceeds.

Aurora-Lens holds the response until the referent is established by the caller — not inferred by the model.

Contradicted fact The AI incorporates a new assertion that contradicts something established earlier in the session.

The model, lacking persistent state, accepts the new input and proceeds.

Aurora-Lens tracks committed state across turns. The contradiction is flagged before the new claim is incorporated.

Governance layer

Evaluates the turn. Decides whether generation may proceed.

↓  permitted only if admissible
Expression layer

Your language model — any provider, any version.

↓  governed output only
Application

Receives output that has passed the gate.

The model produces text.
Aurora decides whether it may.

The language model never holds decision authority. Swap providers, update versions, change fine-tunes — the governance guarantees remain in place.

Licensing, evaluation, acquisition.

If you are evaluating AI governance infrastructure or considering IP acquisition, contact directly. Technical documentation and a working implementation are available under NDA.

Contact Margaret Stokes →
Available Licensing · Acquisition · Evaluation under NDA