See it work, before you talk to anyone.
Four steps on the live system, about 5 minutes, no account and nothing to install. It runs on the real registry — 58 AI systems and 6 policies today — not on a sandbox built to look good.
The walkthrough
Score a document yourself →
Paste any AI system’s published documentation — a model card, a usage policy, a safety page. You get the seven governance questions, which ones it answers, and which it is silent on.
Runs in your browser. Nothing is uploaded, because the page has nowhere to upload it to.
Run a policy decision →
Pick a policy and a model in the playground, send a request, and watch the decision: allowed, redacted, rerouted to an approved model, held for a human, or blocked — with the reason.
The playground is simulated: the router has never carried production traffic, and every recorded request is a test or a refusal.
Verify the evidence, without trusting us →
Each decision is hash-chained to the one before it. The page shows the chain’s status, and the command to check it yourself is on the page.
Tampering with a record breaks the chain, and the break is visible to anyone who runs the check.
Watch a document change →
Vendor documents are fetched and fingerprinted every day. When a page is quietly rewritten, both versions stay on the record with their dates.
This is the part that compounds: the history cannot be backfilled by anyone starting later.
What this walkthrough cannot show you
A tenant of your own — your policies, your keys, your users, and the examination packs built from your own evidence — sits behind a login, and named findings about specific vendors are withheld from public pages until media liability cover is bound. Both are worth fifteen minutes with the person who built this rather than a screenshot.