Primary metric — action-based. Billed per action processed through Agent OS. Risk-weighted: Low 1× · Medium 3× · High 10×. Policy evaluation adds minimal overhead (on the order of tens of milliseconds) and no extra LLM tokens.
Base commitment — per agent seat. A seat is one governed agent instance. Tiers scale 1 / 3 / 5 / unlimited agents and define your monthly action allowance (monthly allowance + overage, with a configurable cap).
Enterprise. Custom action pool per contract, volume discounts, SLA and priority support, optional self-hosted deployment.
Annual plans — Same effective monthly rate with a flexible yearly action pool (e.g. Pro annual: 180,000 actions to draw across the year, not only 15k every month). Auto-charge overages to a user-defined cap — then hard stop.
Token overhead — Agent OS is not a second LLM. Deterministic policy matching + compact provenance JSON; net token increase typically under 5% vs native LLM alone. See How it works → Token overhead.
Read-only. No state change. Safe to replay.
State-modifying but reversible. Staging environments.
Destructive or production operations. May require human approval.
Risk levels use default rules. Team and above can configure custom classifications. Human approval requests consume one action credit on submission.
| Tier | Included actions | Overage per 1,000 | Auto-charge cap |
|---|---|---|---|
| Free | 300 / month | — upgrade prompt | Hard stop |
| Pro | 15,000 / month | €18 | User-defined |
| Team | 30,000 / month | €15 | User-defined |
| Business | 100,000 / month | €10 | User-defined |
| Enterprise | Custom pool | Negotiated | Negotiated |