Three prices, and what none of them includes.
The registry, the method, the corrections log and 7 governance artifacts stay free to read, because a score nobody can re-derive is worth nothing. What an organisation pays for is a tenant of its own: its policies applied to its AI traffic, and the evidence that comes out. Three tiers, and the limits of each are on this page.
Your clients ask what the vendor promised. This is the record.
A model card or a safety page is not evidence a client can rely on until someone has read it, scored what it actually says, and kept a dated copy. A tenant of your own turns that reading — and your client’s own AI-use policy — into a record with a source and a hash behind every line, rather than a screenshot.
- A console tenant: your policies, your keys, your users
- The policy router, with the hash-chained evidence log and exports
- Examination packs you can hand to a committee or a regulator
- The document corpus and its change history, with sources and dates
- Everything in Practitioner
- A workspace per client, delegated from your own tenant
- Exports carrying your firm’s name, not ours
- Two-factor sign-in you can require for every user in your org — off until you turn it on
- Everything in Firm, with terms written for your engagement
- Named findings under engagement, once media liability cover is bound
- Security questionnaires answered straight — ask what we do not yet have and you get a list, not a dodge
Paying starts a subscription; access is then granted by hand in the console against the payment record, usually the same day — nothing is provisioned automatically, and an entitlement cannot be granted without a note saying how it was settled. Prices as of 23 September 2026. Every tier runs on the same router and the same evidence chain: a tier changes settings, never the code that produces a score. The comparable products publish no prices at all — these are ours, in public, for the same reason the method is.
- The registry, the scores, the evidence tiers and every source
- The published method, so any score can be re-derived
- The corrections log, including reviews that changed nothing
- A right of reply for any organisation named, at no cost
- The 7 free artifacts in the marketplace
No assessed vendor can buy anything here — not a score, not a listing, not a faster assessment. Nothing sold on this page moves a review up the queue.
The Practitioner and Firm tiers are billed through Stripe Checkout — Stripe handles the card, not this site. Access is then granted by hand in the console against the payment record, usually the same day: an entitlement cannot be granted without a note saying how it was settled, the same auditability the rest of the system runs on. Enterprise is still settled by invoice outside Stripe.
What that price does not include today
- Production router traffic. The router works end to end and is available to evaluate, but no organisation has yet put live traffic through it. A first deployment is a piece of work we would do together, not a switch to flip.
- Named findings. Which vendor scored what is withheld from public pages until media liability cover is bound. Access under engagement is a separate conversation, not something this price buys.
- Comparative Benchmark Medians — subscriber report. Priced but not available: it reports medians across engagements and cannot be published honestly until there are five of them.
- The infrastructure a bank’s questionnaire asks for. No SOC 2 report, no independent penetration test, and a single server rather than a multi-region deployment. Each is a real gap, listed with what it would take in PRODUCTION.md, and none of them is hidden from a buyer who asks.
- A support desk or an SLA. Rtavan AI LLC is founder-led. You get a direct line to the person who built it, not a 24-hour desk, and no tier buys a response-time promise we could not keep.