Tokens are how models bill
A token is a chunk of text — roughly three-quarters of a word. Providers bill by tokens in and tokens out, so an agent that talks a lot costs more.
Why agents need a limit
An agent loops: it plans, calls a tool, reflects, and repeats. Without a cap, a failing task can loop for thousands of tokens. A limit makes the loop stop at a number you choose.
Two limits that matter
- Per-run — a ceiling for a single task.
- Per-day — a ceiling for everything the agent does in 24 hours.
Set them
- Open the agent’s settings and set a token budget.
- Set a daily spend cap in Settings.
- Add a step budget so a task stops itself early.
What a sensible default looks like
i
Start low — a few thousand tokens for routine jobs — and raise it only when a real task needs more.