Pricing
Models, APIs and Runtimes on one balance. Pay per call, no subscription. Every charge shows up in Activities.
Included on every plan
Up to 50% off
AI models
Live discounts on selected language, image and video models. The price on the card is the price you pay.
100 free calls
Every day
Eligible data APIs share a daily free allowance per organization. It resets at 00:00 UTC.
$0
Discovery
Searching the catalog and reading a tool's schema and price never touch your balance.
$0
Sandboxes
Agent sandboxes are free for now. Fair-use caps may apply, and pricing will be announced before it changes.
$0
Failed calls
A model or API call that fails is not charged. Activities shows the status and cost of every call.
Plans that follow your work
Start
Build
Scale
One activity trail
Component
Model work
Token, image, video or audio generation priced by the selected provider model.
Component
Real-world APIs
Usage follows the published pricing of the API capability you call.
Workflow
Agent runs
See component and runtime work together in each traceable execution.
Choose your way in.
Setup puts tools inside your AI app. APIs let your product call one tool directly. Build Agent is for reusable, multi-step work.
SandBase is for builders who need working AI capabilities quickly, and FDEs who need to turn customer requirements into reusable agent workflows.
Use Setup when you want tools inside Codex, Claude, Cursor, or Kiro. Use APIs when your product needs one tool, like search or image generation. Use Build Agent when the job has multiple steps and should be reused.
The Store is where you find reusable agent capabilities: models, APIs, ready-made agents, and skills. You can test them, copy them, install them into Setup, or clone them into Build Agent.
An agent workflow is a reusable job with instructions and tools. It can research a customer, enrich leads, monitor competitors, triage GitHub issues, or run another repeatable process.
You can test it, publish it so apps and AI tools can call it, or schedule it to run automatically. Every execution becomes a run with input, output, trace, and status.
Webhooks notify your system when async model, API, or agent work finishes or fails, so nothing has to poll for the result.
