Pricing
Pricing is set per engagement — the fastest way to a number is a short conversation about your team and repos. Under the hood it’s simple: licensed seats plus metered model usage, with a monthly budget you control.
For a single team adopting governed AI development.
- Full I3R lifecycle + human gates
- Proposal review with cross-check
- Event log & audit views
- Usage metering & live spend view
- Email/password + Google sign-in
- Up to 10 members
For organisations standardising on agent governance.
- Everything in Team
- Per-tenant SSO (SAML/OIDC)
- Tenant + member approval workflows
- Database design across 5 datastores
- Component catalog & adoption
- Continuous-learning decision KB
For regulated clients with non-negotiable controls.
- Everything in Business
- Non-overridable tenant-floor policy
- Dedicated isolation options
- Compliance onboarding & audit support
- Priority support
How billing works
Seats for your people, metered usage for the AI work — measured where it runs, governed like everything else.
You license people, not API keys. Every AI action traces back to a licensed person — the same attribution your audit trail runs on.
Every model call is counted at the source and rolled up by model and task class — a live month-to-date spend view, not a surprise at invoice time.
Set a monthly budget and get alerts at 80% and 100%. Alert-only by default; flip on enforcement and new agent work stops at the cap.
Tasks route to the most cost-effective model, and bulk work uses batch mode and prompt caching — with the discount visible in your spend view.
Questions teams actually ask
How does billing actually work?+
Two parts: licensed seats for your people, plus metered model usage for the AI work itself. Every model call is counted and visible by model and task class, month to date. You set a monthly budget — the platform alerts your admins at 80% and 100%, and if you opt in to enforcement, new agent work is blocked at the cap and recorded as a budget stop in the audit trail. No external payment processor sits in the loop.
Where does the AI agent actually run?+
On infrastructure you control — the agent daemon runs where your repositories and the Claude Code CLI live, and talks to the platform through a command queue. Your code never has to leave your machines to be worked on.
Can the agent merge to our main branch?+
No. Approved proposals are applied to a dedicated branch and opened as a pull request. The final merge is always performed by your team, in your repository host.
What happens if the agent misreports what it did?+
Every proposal carries both the actual diff and the agent’s own action log, captured independently. If they disagree — an unlogged change, or a claimed change that isn’t there — the proposal is blocked and flagged. It cannot be approved past a failed cross-check.
Do you train on our code or share data across customers?+
No. All data is tenant-scoped. Cross-tenant learning is off by design and would only ever exist behind an explicit, consent-based model.
Which identity providers do you support?+
Any SAML or OIDC provider — Google Cloud Identity, Okta, Azure AD and others — configured per tenant with DNS-verified email domains. Personal email accounts are rejected.