---
title: "Subagents"
description: "Let several agent instances work on independent parts at once."
lang: en
---

A single agent works in sequence, which wastes time on parts that could run together. Parallel subtasks let them start at the same time.

## How it works

The main agent hands part of the work to instances it spawns, keeps working on the rest, and then merges the results.

## Good fits

- A task that splits into independent pieces, such as researching several libraries at once
- Several separate deliverables, such as one summary per module
- A broad search where splitting clearly shortens the wait

## Keep in mind

Every parallel instance spends model quota of its own. Each works within its own scope; the main agent merges the results before delivering, rather than dumping them on you.

## Related

- [Sessions](/docs/using-bi/sessions)
- [What it can do](/docs/using-bi/tools)