Priced on volume, not seats.

Free

25,000

Full enforcement and signed evidence, in production, with nothing held back.

$0forever
runs / month
25,000

runs / month

30 days

Debug what happened last month.

  • Full enforcement - all 13 rule types, all 19 PII types
  • Allow, block, redact, and review verdicts
  • Prompt-injection scanning
  • HMAC chain signing + server countersignature
  • Offline verifier
  • All 30 integrations
  • Basic event list
  • Recording pauses at 25,000 runs, enforcement keeps running
  • Docs and GitHub issues
Request early access
Most popular

Pro

500,000

For a team running a real AI product, heading into an audit.

$249/ month
runs / month
500,000

runs / month

13 months

Prove a full audit window.

  • Everything in Free, security is identical
  • Investigation console: replay, incidents, event studies, alerting
  • 1-click evidence export (SOC 2, EU AI Act packages)
  • HITL approval workflows
  • Framework mapping views
  • Email support, same-day response
Request early access

Enterprise

Unmetered

Custom volume, your perimeter, procurement handled.

Contact us
runs / month
Unmetered

runs / month

Custom

Whatever your regulator asks for.

  • Everything in Pro
  • VPC and self-hosted deployment
  • Gateway deployed and operated inside your VPC
  • SSO and SIEM export
  • DPA, security review, procurement support
  • Dedicated Slack channel, priority response
  • Dedicated onboarding
Talk to us

If you go over

$0.50 per 1,000 runs

The same rate you already pay inside the plan. Growing carries no penalty here.

Free
Recording pauses. Enforcement continues. Upgrade to resume sealing.
Pro
Nothing pauses. $50 per 100,000 runs, the same rate as in-plan.
Enterprise
Contractual, agreed up front.

What counts as a run

One run is one governed unit of work: a direct model call, or one agent invocation, however many steps it takes.

  • Blocked runs count, because the block is the work and the record of it is the evidence you are buying.
  • Retries count as separate runs.
  • An agent invocation is one run no matter how many steps or tool calls it makes internally. We meter decisions you are accountable for, not infrastructure.
  • Runs are counted per calendar month and reset at the start of the next one.

Pricing questions

7 questions