Free
$0
Try it. No credit card.
- Evaluations / month
- 100 / month
- Token ceiling
- Small
- Audit retention
- 7 days
- BYOK
- WorkOS SSO
Pricing
Pay per evaluation.
Not per token.
One evaluation = one policy invocation or one summarise call. A 4-step chain = 4 evaluations. You pay for the work the engine actually does, not the tokens that flow underneath — token usage is the abuse cap, not the meter.
20% annual discount on Starter, Pro, and Team · Sellable expansions in 25 / 50 / 100% increments
Starter
Contact us
Single team in production. Predictable monthly volume.
- Evaluations / month
- Volume tier 1
- Token ceiling
- Medium
- Audit retention
- 30 days
- BYOK
- WorkOS SSO
Pro
Contact us
Heavy single-team or light multi-team use.
- Evaluations / month
- Volume tier 2
- Token ceiling
- Large
- Audit retention
- 90 days
- BYOK
- WorkOS SSO
Team
Contact us
Multi-team. BYOK + SSO unlocked.
- Evaluations / month
- Volume tier 3
- Token ceiling
- X-Large
- Audit retention
- 180 days
- BYOK
- WorkOS SSO
Enterprise
Custom
Org-wide. Custom contract.
- Evaluations / month
- Custom
- Token ceiling
- Custom
- Audit retention
- Custom
- BYOK
- WorkOS SSO
How billing works
One billable unit: the evaluation
An evaluation is one policy invocation or one summarise call. Concretely:
- 1
/summarisecall = 1 evaluation - 1 single-policy
/policies/evaluate= 1 evaluation - 1 chain of 4 policies = 4 evaluations
- 1 graph with 6 nodes (2 skipped via conditional edges) = 4 evaluations
Skipped graph nodes don't count — you only pay for the work the engine actually performs. Includes validate-mode (pure DSL) because it's real work with a real audit trail, even without an LLM hop.
Token ceiling = abuse cap, not the meter
Each tier carries a token ceiling on top of the evaluation count. It exists to catch pathological workloads (a single evaluation running thousands of tokens of unnecessary context). Most legitimate workloads never hit it. Sellable in +25 / +50 / +100% expansions when you genuinely need more headroom.
Expansions, not overages
Hit your evaluation cap mid-month? Buy a +25 / +50 / +100% expansion as a standard SKU. One-time (this period) or recurring (auto-renews). We don't do mid-month overage billing — expansions are predictable and procurement-friendly.
BYOK on Team+ and Enterprise
Bring your own LLM provider keys (Anthropic, OpenAI, Groq, Mistral, or any OpenAI-compatible endpoint). You pay the provider direct; we charge for the orchestration value. Required for regulated workloads where you need to own the LLM relationship.
Transparency built in
- Token counts come straight from the provider — never estimated.
- Cost computed at call time using the active rate card. Historical invoices stay reproducible even if rates change.
- Per-evaluation drill-down in the admin: every LLM call, model, tokens, latency, cost.
- Self-serve CSV download of every billable evaluation.
Quota enforcement
When your tenant reaches its effective evaluation or token cap, the API returns
429 quota_exceeded with the cap, the current usage, and the upgrade or expansion options. At 80% of either cap we add an X-Usage-Warning response header so you can react before hitting the wall.FAQ
Why no dollar amounts on the page?
Pricing is set per-engagement during the design-partner phase. We'll publish flat tier dollars once we have validated workload data across production tenants. Request a demo — we'll quote based on your volume.
What happens at the evaluation cap?
The API returns a 429 with
{ error: 'quota_exceeded', used, cap, kind } where kind is evaluation or token. Options: buy an expansion, upgrade tier, or wait for next month. Nothing surprises your bill.Doesn't per-evaluation billing penalise chains and graphs?
We considered per-workflow (one top-level API call = one unit regardless of internal steps) and concluded it didn't reflect cost — a 4-step chain consumes 4 LLM calls and costs us ~4× more than a single evaluation. Per-evaluation aligns billing with cost-of-goods. Composition is still the differentiator; flattening chains into mega-policies uses MORE tokens, hits the token ceiling sooner, and loses per-step audit granularity — so operators have their own reasons to keep workflows composable.
Can I downgrade tiers mid-month?
Yes. Evaluations you've already run stay attributed to the tier that was active when they ran. New evaluations count against the new tier from the moment the change applies. We don't retroactively re-bill.
What about Healthcare or EU customers?
DPA available on Team and contractually included on Enterprise. BAA on Enterprise only. EU data residency on the engine's roadmap — talk to us about timing.
Is the engine LLM-agnostic?
Yes — every tenant routes to their own provider/model under Engine Settings → Model routing. We support Anthropic, OpenAI, Groq, local Mistral, and any OpenAI-compatible endpoint. BYOK customers use their own keys; managed customers use ours.
Ready to talk numbers?
Tell us your expected volume and compliance requirements. We'll quote a tier and timeline.