Start free. Scale when the automation proves its value.
Every plan includes shadow mode, policy enforcement, run evidence and audit history. Higher plans add volume, enterprise identity and stronger isolation.
Build one automation and run it against your own systems.
- Every guardrail, including the forbidden list
- Full audit trail and run traces
- The build agent, against your own schema
- Runs
- 100 per month
- Model spend
- $5 per month
- Build turns
- 100 per month
- Live automations
- 1
- Dedicated database
- Pooled with row-level security
Several automations in production, with approvals and audit.
- Approvals with business roles
- Scheduled and webhook triggers
- Boundlane Exchange, the template marketplace
- Runs
- 2,000 per month
- Model spend
- $50 per month
- Build turns
- 500 per month
- Live automations
- 10
- Dedicated database
- Pooled with row-level security
Production volume, SSO, and the option of a dedicated database.
- SSO via OIDC or SAML, and SCIM provisioning
- Optional dedicated database
- Externally anchored audit chain
- Runs
- 20,000 per month
- Model spend
- $300 per month
- Build turns
- 3,000 per month
- Live automations
- 50
- Dedicated database
- Available
On these plans, model spend is included in the price and enforced as a ceiling, not billed on top. At the limit, builds and runs stop rather than continuing onto an invoice — no usage is metered, and nothing is passed through at cost. Enterprise is the plan where they do not stop.
Company-wide governance, without a ceiling.
Team and Business stop at their limits. Enterprise has none to stop at: usage is measured and settled against a negotiated agreement rather than capped by a counter. For a process that releases payments or closes a period, not stopping is the reason to be on it.
With it: a dedicated database, your own model endpoint, negotiated support terms, and deployment in your own cloud account by arrangement. Priced against volume and isolation rather than seats.
Every number above is read from the plan table itself rather than transcribed beside it. Runs, model spend and build turns are the three the runtime checks before it starts work, and they cannot drift from this page, because there is only one table.
Questions worth answering first
Work stops. A run or a build turn is refused with a quota error rather than run and charged for, the counters reset at the start of the next calendar month, and moving up a plan raises the ceiling immediately. Enterprise is the exception, and deliberately so: stopping a payment run because a counter tripped is a worse outcome than a conversation about an invoice.
The platform and the model spend both. There is no separate model bill, no seat count and no charge for a run that was refused. What the price does not include is more than the plan says: the model-spend figure is the most a workspace can consume, not a starting balance.
Because it is the cost that varies with what your automations do rather than how many you have. A run that reads three records and one that reads three thousand are the same run and very different money, so both are counted.
No. A deployment on Vercel authenticates to the AI Gateway with the token the platform injects. Bring your own model endpoint on Enterprise if your data cannot leave a boundary you control.
One message to the build agent and everything it does in response — reading your schema, editing the package, running the promotion gate. Building an automation typically takes a handful; refining one takes a few more.
On Business and Enterprise, your data can live in its own dedicated database rather than a shared one, and Enterprise can bring its own model endpoint. Full deployment into your own cloud account is the one step beyond that: the tooling exists — an install can be planned and verified from outside — and what we have not agreed yet is the support model for software running where we cannot see it. Talk to us if you need it.