The assistant
Task token budget
The assistant · 1 min read
Optionally sets a hard ceiling for one assistant task, checked before each new request can spend your key.
When you would want it
Long tool loops and repeated context can cost far more than the first question suggests. Checking the whole request before it starts is safer than warning after the tokens are gone.
Where to find it
- All Features → Task token budget
- Settings
What you get out of it
- A hard pre-request ceiling beats any after-the-fact warning label.
- Runaway tool loops stop on their own instead of draining your key.
How to use it, step by step
- Paused while the assistant edits directly: with no isolated tasks created, there is nothing to cap.
- Unlimited stays the default; the rows are disabled until caps return.
- Runaway tool loops are still bounded by the per-turn step limit instead of draining your key.
Why this beats the usual way
Most tools warn after tokens are gone, or not at all. Reserving before the request is the only version of the feature that cannot be circumvented by a bad loop.
Find it in Universal Search
Task token budget · The assistant