Usage-based. No subscriptions, no seats. Five spend categories, same as your billing — passed through with a small operating margin baked in.
Compute the agent uses to think, train, and run your Machines. The biggest line item for most teams — but the one with the most room to optimize.
Reasoning + tool calls. Priced per token, in and out.
Image, vision, and other multimodal outputs.
Web search, citation pulldowns, and external API calls.
Persistent data, embeddings, and Machine state.
Numbers above are typical-workload examples. Your billing page shows the canonical unit price for each category at runtime.
Ad hoc analysis, a couple of quick predictions a week.
A single trained model scoring leads or churn risk daily.
A few Machines + interfaces shared across a small team.
most teamsVision, large embeddings, multiple scheduled forecasts.
Priority response, sync calls, custom onboarding, dedicated channel. Flat monthly. Cancel anytime.