Credits & billing
Credits are how usage is metered. A credit is a small fixed slice of real model cost; every task spends credits on model tokens plus any external tools it calls (search, image/video, browser, sandbox). Each organization gets free credits every month that reset; paid plans and one-time top-up packs add more. See current pricing on the Billing page.
What a credit is
A credit is a fixed slice of the underlying model and tool cost, billed per organization. Tasks consume credits two ways: language-model tokens (input + output, priced per model) and external tool calls that cost real money to run.
Cheaper models cost far fewer credits per token than premium ones, so your per-task burn depends mostly on which models your agents use.
Free credits, plans & top-ups
Every organization gets a monthly credit allotment that resets each billing period (unused plan credits don’t roll over). Paid plans raise the monthly allotment. You can also buy one-time credit top-ups at any time — by the pack or by the amount, with or without a subscription; top-up credits stack on your balance, are used after the plan’s credits and never expire.
| Plan | Monthly credits |
|---|---|
| Free | 2,500 |
| Lite | 30,000 |
| Starter | 100,000 |
| Scale | 550,000 |
| Agency | 1,200,000 |
What consumes credits
Most spend is model tokens. On top of that, some tools have a real per-call cost (shown below in USD-equivalent so you can compare). The agent only calls a tool when the plan needs it.
| Action | Approx. cost per call |
|---|---|
| Generate image | $0.04 |
| Generate video | $0.03–$0.78 per second of video, by model and resolution (a 5-second clip ≈ $0.15–$4) |
| Web search | $0.01 |
| Crawl a page (scrape) | $0.01 |
| Browser session | $0.05 |
| Python sandbox run | $0.002 |
| MCP tool call from your own agent (tools mode) | $0.001 per read, $0.005 per write — plus a paid platform tool's own cost |
See your usage
The Billing page shows, for the current organization:
- Your remaining credit balance for this period.
- Usage over the last 7 and 30 days, with input/output token counts.
- Credits by model — which models drove your spend.
- Self-serve credit top-ups (packs or a custom amount) and what each plan includes — credits per month, estimated tasks, parallel tasks — with your current plan highlighted.
Lower your credit burn
- Set cheaper models per agent on the Agents page — specialists default to a fast, low-cost model while the orchestrator stays on a quality planner.
- Scope tighter date ranges and fewer accounts when you don’t need a full history.
- Pin a Skill so the agent follows a proven, shorter plan instead of re-deriving one.
- Prefer read-only audits first, then run only the writes you actually want.
- Already paying for your own model deployment? Bring your own model zero-rates LLM tokens entirely — only tool/media calls still meter.
Frequently asked questions
- Do unused credits roll over?
- Plan credits reset each billing period and don’t roll over. One-time top-up packs stack on your balance and do not expire at reset.
- What happens when I run out?
- Runs are blocked at zero credits until the next monthly reset, a plan upgrade, or a top-up pack. Nothing is charged silently.
- Is billing per user or per organization?
- Per organization. Each org has its own balance, plan, usage history and invoices.
- How do I buy more credits?
- On the Billing page: subscribe or upgrade your plan for a larger monthly allotment (the cheapest credits), or buy a one-time top-up — a pack or any amount — via self-serve Stripe checkout, with or without a subscription.