Delegation

Some jobs are too big for a single plan — crawling dozens of pages, processing many items one by one. The agent splits these across sub-agents with delegate_task and merges the results.

How it works

• The main agent breaks the job into pieces and hands each to a sub-agent.
• Independent pieces run in parallel; total time drops to the slowest piece, not the sum of all pieces.
• Sub-agent results are collected, and the main agent synthesizes them into a single answer for you.

Progress messages

On long delegations, to avoid any "is it stuck?" worry, the agent returns progress messages (e.g. "12/40 items processed"). You get partial visibility before the task finishes.

Security & limits

Sub-agents pass through the same policy engine: each is subject to allow-lists, risk classes and approval thresholds on its own tool calls. Delegation speeds work up but does not relax the security model.