Autonomy levels
Every agent has an autonomy level that controls how independently it can work before pausing. Think of it as a dial between “always check with me” and “get it done.”
You can change an agent’s autonomy level from its detail page in the Selu web interface.
The three levels
Section titled “The three levels”Best for everyday tasks where you want to stay in the loop.
- Up to 24 tool steps per turn
- Up to 1 specialist handoff (when your agent passes a task to another agent)
Use this when you want the agent to do simple, bounded tasks and stop if anything unexpected comes up.
Medium
Section titled “Medium”Good for multi-step workflows that need a bit more room to breathe.
- Up to 64 tool steps per turn
- Up to 5 specialist handoffs
This is a solid default for most users. The agent can handle more complex tasks — like researching a topic across several sources, or coordinating between a couple of specialist agents — without going off the rails.
For long-running or deeply automated workflows, like coding tasks or complex multi-agent pipelines.
- Up to 200 tool steps per turn
- Up to 6 specialist handoffs
Why is there a cap at all?
Section titled “Why is there a cap at all?”Every level has a finite step limit. This is intentional — it’s a safety net that prevents your agent from getting stuck in a loop and running up costs if something goes wrong with a request.
If your agent regularly hits its step limit, that’s usually a sign the task is too broad, or that the agent needs more specific instructions rather than more autonomy.
Specialist handoffs
Section titled “Specialist handoffs”A specialist handoff is when your agent delegates part of a task to another agent. For example, a general assistant might hand off a coding subtask to a dedicated coding agent.
Each handoff counts toward the limit for that turn. This keeps multi-agent workflows predictable and prevents chains of agents from running indefinitely.