Concept

Limits and billing

What your agent's calls cost, the rate limits per token and per business, and what to do when a call is told to slow down.

Last updated Open in ClaudeOpen in ChatGPT
On this page

Your own agent does the thinking when it works on Willo, so Willo charges nothing for its reads and changes. The limits below protect your business and keep Willo responsive for everyone.

What costs credits

ActionCredits
Any read: files, backend, logs, the businessNone
Any change: files, tables, rows, functions, contentNone
Site rebuilds and publishes your agent triggersNone
willo_task, which runs a full Willo chat turnBilled as one chat turn
willo_docsNone

Connecting agents at all comes with the Builder plan and up; below that, tokens can't be created and existing ones are refused. See Pricing.

A task bills exactly like a message you type in Willo's chat, and needs the business to have credits. See Handing Willo a task.

Rate limits

LimitValue
Calls per token per minute60
Calls running at once per token6
Calls per business per minute, all agents together300
Changes per business per minute, all agents together30
Site file commits per business per hour40

The full list, including the metered kinds of call, is on Limits.

Metered calls

Some kinds of call, such as reading or committing your site's code, backend management and Stripe reads, have their own per-minute limit for each business. Heavy use in one business never slows another. Tools that count toward one are marked Metered in the tool reference.

When a call is told to slow down

CodeWhat happenedWhat to do
rate_limitedThe token or the business made too many calls this minute.Wait for the time in the message, then retry.
write_cappedThe business took 30 changes this minute.Batch changes, then retry.
budget_exceededThe business reached the per-minute limit for one metered kind of call.Wait, then retry with fewer calls.
too_many_cardsThe token has five approval cards waiting.Ask the person to answer them first.

Each message says how long to wait. Nothing is lost: a refused call changed nothing.

Making fewer calls

  • Change several parts of one file in a single willo_edit_file call: it takes up to 25 edits, applied together or not at all.
  • Read with a window. willo_read_file returns 500 lines by default; ask for the part you need.
  • Describe one table with willo_db_schema rather than listing everything each time.
  • For a large multi-step job, one willo_task may be simpler than dozens of direct calls, at the cost of a chat turn.

Approvals have time limits too

A card stays open for seven days. After a person approves, the agent has one hour to retry the call. See Approvals.