Set exactly how many actions one employee can take while working on a single task, from 50 up to 1,000. Give a research-heavy employee more room to search, read, and delegate before it has to stop and report back, or cap a simple support employee tighter so it never wanders past what the task needs.
Every AI employee works inside a budget: a maximum number of tool calls and reasoning steps it can take before it has to stop and hand a result back to you. By default that budget comes from your workspace's own setting, which itself falls back to a system default sized for real production work. Set a Per-Task Action Limit lets you override that number for one specific employee, from the employee's profile, without touching the workspace default or any other employee's setting.
You will find the control in the Brain section of an employee's Profile tab, next to model and reasoning-depth selection. Leave it blank and the employee inherits your workspace's action limit; set a number and that employee always uses it, task after task, until you change it back. The value is a single integer between 50 and 1,000: low enough to bound a narrow employee's cost, high enough to cover a leader running deep multi-step delegation.
This is different from picking a faster or slower model. Model choice changes how the employee thinks on each step; the action limit changes how many steps it gets before the turn ends. A researcher pulling from a dozen sources, cross-checking each one, and then writing a summary needs a high ceiling, or it gets cut off mid-task with an incomplete answer. A single-purpose support employee replying to one ticket rarely needs more than a fraction of that, and a tighter limit keeps its cost and behavior predictable without limiting what it is capable of in principle.
The setting is per employee, so a team of five can each carry a different ceiling that matches what they actually do. Nothing else about the employee's skills, tools, or personality changes: this is purely the size of the box one employee's work happens inside.
Open the employee's Profile tab and look in the Brain section, alongside the employee's model and reasoning-effort settings. The action limit sits next to those because all three shape how the employee spends effort on a task: which model it thinks with, how deeply it reasons on each step, and how many steps it gets in total.
Type a number between 50 and 1,000, or clear the field to go back to inheriting your workspace default. The change applies the next time that employee starts a task; work already in progress keeps running under the limit it started with.
Start from what the employee actually does. An employee that searches the web, reads several pages, cross-references facts, and writes a long report needs a high ceiling, closer to the 1,000 maximum, because each search, each read, and each delegation to a teammate counts as a step. An employee handling one narrow, repeatable job, like drafting a single email or updating one record, rarely needs more than a couple hundred.
If you are unsure, leave it inherited and watch how the employee's real tasks behave. If you consistently see a task get cut off before finishing, raise the number for that employee specifically rather than raising the whole workspace default, which would loosen the ceiling for every employee at once.
One number, resolved per employee, per task
Every employee turn runs as a loop: think, call a tool, read the result, think again, until the task is done or the loop hits its ceiling. Set a Per-Task Action Limit is the number that defines that ceiling for one employee, expressed as the maximum number of loop steps it is allowed to take in a single task before it must stop.
When you leave the field blank, the employee falls back to your workspace's own default, and if your workspace has never set one, it falls back to the system default. Setting a number on the employee overrides both, and it stays in effect for every future task that employee runs until you clear it or change it again. There is no separate toggle to turn the override on: entering a number turns it on, and clearing the field turns it back to inherited.
If an employee reaches its limit mid-task, it stops there and reports what it has so far rather than continuing indefinitely. That protects your workspace from a broken loop running away unbounded, while still giving you the room to raise the ceiling for employees whose real work genuinely needs more steps.
A researcher pulling from a dozen sources, cross-checking claims, and drafting a long report needs many tool calls in one task. Raise its limit toward the maximum so it completes the full task instead of stopping mid-research.
A support employee that answers one ticket at a time rarely needs hundreds of steps. Set a lower ceiling on that employee so its behavior and cost per task stay consistent and easy to reason about.
A leader that plans work and delegates pieces to several teammates uses steps for its own reasoning plus every delegation round trip. A higher limit on the leader specifically prevents a busy sprint from getting cut off before every teammate reports back.
If one employee's real tasks keep hitting the workspace default, raise the limit on that employee alone instead of raising it workspace-wide, so every other employee keeps the tighter, more predictable ceiling.
Each step in the employee's think-act-observe loop counts: calling a tool, reading a file, running a skill, or delegating to a teammate. A task that only needs a couple of tool calls and a final reply uses very few of its budget; a task that searches, reads, and iterates many times uses far more.
The employee stops the current task at that point and reports back with whatever it completed so far, instead of continuing indefinitely. Raising the limit for that employee gives it more room to finish similar tasks in one pass next time.
No. It only affects how many steps the employee gets in a single task, not the model it thinks with or the quality of any individual step. Model and reasoning-effort settings, set separately in the same Brain section, control how the employee reasons; this setting controls how long it is allowed to keep working before it has to stop.
Yes. The override lives on the individual employee, so a research-focused employee can carry a high ceiling while a narrow, single-purpose employee stays capped lower, all inside the same workspace.
The employee inherits your workspace's own action limit, and if your workspace has never set one, it falls back to a system default sized for real multi-step production work, including delegation to teammates.
Set a Per-Task Action Limit is part of Things that control how they behave.
Skip the settings panels. Tell your AI employee what skills to learn, what rules to follow, and what personality to use, all through natural conversation. They can even browse the skill catalog and configure themselves as work evolves.