Cultural authorization for AI agents.
TastePolicy
Permission is not preference.
AI agents are learning how to act for us before they've learned how to have our taste. TastePolicy checks subjective choices against a human's configured cultural policy before they become actions.
Inspect the working demoWorking demo / SELECT_ENTITY
From signals to an inspectable decision.
Verified M1 foundation · Experimental policy
Optional versioned state; no LLM or action execution.
Judge Demo
Your AI wants to choose a place for you. Should it?
Load five reviewed concepts and the place target. This resets the current demo; real resolution and your explicit identity choices still come next. No request runs automatically.
Your taste signals
Prefilled names are concepts, not identities verified in this session. Resolve each, then explicitly choose its identity. Review Perfect Days (2023), Piranesi by Susanna Clarke and the TV show Severance; ambiguous matches stay visible.
Cross-domain evidence
The server re-verifies entity IDs. Rank is Qloo's returned order, not a policy decision.
0/5 identities selected. Five distinct verified signals are required.
No candidates yet. Real Qloo evidence is required.
Cultural policy evaluation
A proposal needs evidence.
Resolve your signals and discover candidates in place. Then propose one specific entity.
No decision issued.
A separate candidate-specific Qloo query supplies evaluation evidence. Discovery rank does not authorize an action.
Cultural Authorization Receipt
A captured audit snapshot of the evidence and policy used for this choice.
Evidence before authorization.
A receipt appears only after successful provider evidence evaluation. Unavailable evaluations do not manufacture a receipt.
Versioned Taste State
Temporary pseudonymous profile · 24-hour expiry. Explicit feedback belongs to the subject; agent labels do not create separate profiles. No learning or confidence claim.