Pricing
People are free. You pay for the agents you put behind a key
Your teammates approve, review and export as much as they like on every plan. The meter runs on gated agent identities and on calls. Going over your plan starts a trial and sends you a note instead of stopping production.
Free
One person, real keys, the first agents
$0
forever
Start free- 10,000 calls a month
- Up to 5 gated agents
- Unlimited people, approvers and reviewers
- MCP, HTTP proxy and SSH
- Approvals in the console and in Telegram
- Hash-chained journal with signed checkpoints
- Journal export and the open verifier
Team
A team whose agents touch production
$19
per gated agent, per month
Start free- 1,000,000 calls a month
- Unlimited gated agents
- Policy profiles across many bindings at once
- Observe mode and the simulation report
- Time-boxed grants with call budgets
- Workspaces, roles and delegation
- Webhooks and the full REST API
Business
Agencies running agents on other people's systems
$99
per client workspace, per month
Talk to us- No call limit
- A separate vault, approver group and journal per client
- Guest approvers by magic link, no account needed
- Cards branded as yours
- A per-client report and journal segment they can keep
- Workspace cloning when you sign the next one
- A dedicated deployment, on your hardware or in your region
Prices are per month and exclude tax. A gated agent is one agent identity that holds bindings; the people who approve its calls never count against it.
Pricing
The limits, and why each one exists
A limit lives here only when removing it breaks memory, money or somebody else's service. Each row carries its reason in the same line.
| Limit | Value | What breaks without it |
|---|---|---|
| Calls on your plan | 10,000 free · 1,000,000 team · no limit on business | Billing. Going over starts a trial and sends a note. Calls keep running |
| Rate per hour | None for the first 30 days, then 10,000 | Somebody else's API, and your bill. Answers with a retry time and never raises a card |
| Calls at once | 32 per agent · 256 per tenant | The executor's connection pool |
| Inline answer | 1 MiB per call, per binding | The model's context. The whole body stays available by reference |
| Upstream body | 512 MiB | Executor memory and object storage |
| Pending approvals | 20 per agent · 100 per tenant | A storm of cards. Over the line the agent pauses with one card instead of a hundred |
| Read budget | Five times your own hourly median, learned over seven days | One compromised agent dumping every SaaS you have. Crossing it raises a single summary card |
What people ask about the price
- What counts as a call?
- One action your agent asks for and the gateway executes: one tool call, one proxied request, one SSH command. Retries after a refusal do not count, and neither does listing the tools.
- What happens when I go over?
- A trial starts and you get a note. Your calls keep running. Nothing about billing ever raises an approval card or blocks a production request.
- Do approvers cost anything?
- No. People are free on every plan, including the client-side approvers on Business and the guests who answer by magic link.
- Can I keep the data in my own region?
- Business runs as a dedicated deployment: your hardware or your region, the same binary and the same pipeline. We serve Kazakhstan and Russia today. The compare pages say who else does.
Start on the free plan with a real key
Ten thousand calls a month covers a coding agent working against a real production key. Go over it and you get a note, never a blocked call.
No card. Going over the plan never blocks a call.