Skip to main content

What costs money

Which SDK calls produce billable AI tasks, how manager billing works, how work inside one conversation folds into a single task, and the plan at a behavior level.

The decision: you are building something that runs AI work. What will it bill, and how do you keep it predictable?

What bills​

  • AI execution bills; plumbing does not. Reading a table, listing agents, or sharing a resource is not a billable AI task. Running the models is: sending a message to an agent (autessa.agents.execute) bills agent execution, and holding a manager conversation bills for its work.
  • Manager billing = the configured sub-agent count. A manager conversation bills the number of sub-agents it is configured to run — the sub-agents do not bill separately on top. Choosing how many specialists a manager coordinates is a direct cost decision.
  • Work inside one conversation folds into one task. When an agent or manager turn makes several model/RAG calls to answer, those fold together into that conversation's single billable task rather than each becoming its own charge — so a multi-step answer is metered as one unit of work, not many.

Seeing the bill​

Read-only usage against every limit dimension is available from autessa.company.usageStatus — its AI-task block reports this period's task counts, the resolved per-task rate, and the projected month cost. Invoices, the per-task cost drill-down, payment methods, and commitment changes are billing administration: they live in the browser billing UI, not the script SDK.

The plan, behaviorally​

Billing is pay-as-you-go with an optional monthly commitment: you pay per AI task, and committing to a monthly minimum lowers the per-task rate. There is no free tier. The exact rates and thresholds are shown to the user in the billing UI and the published pricing docs. A script that needs the current rate should read the live figure from the AI-task block of autessa.company.usageStatus rather than hardcode any number, because the rate depends on the company's commitment tier.

Decision rules​

  1. Budget managers by sub-agent count. More specialists means a better answer and a higher per-conversation bill; pick the smallest team that does the job.
  2. Prefer a plain script when no judgement is needed. If deterministic code can do the work, it costs nothing in AI tasks — see the agents-tools-managers guide.
  3. Read rates live; never hardcode prices. Pull the current per-task rate from autessa.company.usageStatus, because rates depend on the company's commitment tier.

Worked example​

A nightly report that runs a manager to summarize the day will bill that manager's configured sub-agent count each night. To estimate the monthly cost, multiply by the schedule and read the per-task rate from the AI-task block of autessa.company.usageStatus.

See also​

  • agents-tools-managers — the units of work that do and don't bill.
  • automations — the schedules and triggers that decide how often billable work runs.