The registry
These are the entries every other view draws on — filter by risk tier or name, and select one to read its full record: scored dimensions, the quoted evidence behind each score, provenance, and any challenges. Each entry carries two independent marks: a risk tier (how risky the agent is) and a trust level (how reviewed the entry is).
Classify an agent
Pick an example for an instant, identical classification — cached for reliability — or paste your own below to classify it live. Every score is proposed by a model, then overridden by deterministic rules wherever a fact can be pinned down, and ratified by an accountable human — new classifications enter the same registry you can browse.
Or try an example
Runtime policy gate
The gate checks agents against the same registry you can browse — at the moment an agent acts — mapping each tool's evidence-backed risk tier to a policy action (allow, allow with logging, or escalate to a human), deterministically, with no model at enforcement time.
Scenario · an orchestrator paying an invoice
An orchestrating agent works through an invoice-payment task; before each step it checks the registry against the bank's policy.
Active policy
Check any agent against the policy
Pick any registry entry to see the decision the gate would return for it.