You pay for the infrastructure, not the tokens.
Pricing is usage-based. You pay a platform rate for everything that makes an agent safe to ship — isolation, memory, the full audit trace, approvals, and channels. Model usage is a separate axis: bring your own key, or let us handle it. No seats, no minimums, and no flat fee — except for enterprise installations.
Get started for free.
Usage-based
Pay only for what your agents do. Billed monthly on usage, anchored on active smiths — your active end-users — so the bill tracks your own growth. No seats, no minimum, no flat fee.
- Get started for free
- Scales down when your smiths are idle
- Every euro itemized in the trace
- Bring your own model keys, or have us manage them
Enterprise
A flat annual contract with a dedicated or self-hosted installation — for teams that need the platform inside their own boundary, with the residency and procurement guarantees their security team signs off on.
- Dedicated or self-hosted installation
- SSO and procurement paperwork
- Custom SLAs and support
- Volume terms on usage
Two things, kept separate.
Platform usage is what we charge for. Model usage is the model provider’s — yours to own directly, or ours to manage.
Platform usage
The infrastructure that makes an agent safe to ship, metered on active smiths as you use it.
- A private, isolated agent per end-user
- Persistent per-smith memory
- The full replayable, costed audit trace
- Human-in-the-loop approvals
- Built-in Slack, WhatsApp, Telegram, and email
- Per-user cost metering and budgets
Bring your own keys
No model markup from usConnect your OpenAI, Anthropic, or Google account. Model usage is billed by them, directly to you. We never sit in your model billing.
Managed model access
Convenience, metered per requestSkip provider accounts entirely. We handle keys, rate limits, and failover, and meter model usage at a service rate over provider prices.
Either way, every request’s exact cost appears in its trace before you’re billed.
We charge for trust, not tokens.
Tokens are a commodity that gets cheaper every quarter. The value is the isolation, the memory, and the record around them — the part that lets you put an agent in front of a customer. That’s what your platform rate buys.
No fine print.
- Do you mark up model tokens?
- Bring your own provider key and no — you pay OpenAI, Anthropic, or Google directly, and we never touch model billing. Use managed model access and model usage is metered at a service rate over the provider's published prices, covering routing, the cost trace, and per-request metering. Either way, every request's exact cost shows in the trace before you're billed.
- Are there seats, tiers, or a monthly fee?
- No — self-serve is usage-based, anchored on active smiths (your active end-users). No per-seat fees, no feature gates, no minimum. The only flat fee is Enterprise: a dedicated or self-hosted installation with custom terms.
- Do I need a credit card to try it?
- Yes. A card activates the account and unlocks your first €10 of credit, on us. It isn't charged for that — you pay only when you top up beyond it.
- Can I see costs before I'm billed?
- Yes — that's the whole point. Every run is costed in its trace, attributed down to the individual smith, and surfaced in the console as you go. No end-of-month surprises.
- What about high volume or procurement?
- Talk to us. Usage-based stays usage-based, but we'll sort out volume terms and the paperwork your procurement team needs.
Start free. Pay for what your agents do.
No seats, no minimums. Every euro shows up in the trace before it shows up on a bill.