Calling another agent

The Call another agent step runs another published agent and uses its result. It is how you compose automations without duplicating logic.

Choosing the target

Studio lists the agents this agent can call, with the name, description, exact version, what the agent asks for as input, and what it returns.

What cannot be called appears with the reason, rather than simply vanishing:

Reason	Meaning
No active version in this environment	The agent exists and is not active here
From another environment	The target has to be in the same environment
The agent itself	Calling itself is direct recursion

The target is always exact

You point at a specific active version, never at "the latest". Publishing a new version of the called agent does not by itself change what this one runs — what happens is that the previous version stops being active.

That is what keeps a composed flow running exactly what was tested.

Mapping the input

The panel shows the fields the called agent expects — with label, type, requiredness, and options — and you connect each one to a value in the current flow. No guessing names.

What it produces

The child run's identifier and the result it returned.

The child run is first class: it has its own timeline and can be followed, cancelled, and audited. In the parent agent's timeline it appears as a step with a link to the full run.

Protections

Protection	How it works
Depth	The agent-calling-agent chain has a ceiling, declared in the agent and capped by the space (default 3)
Recursion	The actual call chain is checked; a cycle A → B → A is blocked
Environment	The child run always runs in the same environment as the parent
Fan-out	One run does not spawn more than 50 children — the same ceiling as "For each item"

The chain of responsibility

In the child run, "who asked" is recorded as the parent run, not as a person. The whole chain is explicit in the audit trail.

When to use it instead of repeating steps

Situation	Why
The same logic appears in several agents	One agent, called by all of them
Handling one list item is complex	That is what "For each item" does internally
Different teams look after different parts	Each publishes and versions its own

Common errors

Symptom	Cause	What to do
The agent I want does not appear	It has no active version in this environment	Activate it in the same environment
Failure because depth was exceeded	The chain is longer than the ceiling	Shorten the chain or revise the agent's ceiling
Failure because of recursion	An agent in the chain calls someone already in it	Redesign the flow
The child run failed	The problem is in the called agent	Open the child run from the timeline

Next steps

• Branching and repetition
• Publishing and environments
• Following runs