并行子智能体
将可并行的任务分派给多个子智能体。
单个智能体按顺序处理任务，遇到本可以同时进行的部分就会浪费时间。并行子任务让它们同时开工。
如何工作
主智能体把其中一部分交给自己派生出的其他实例处理，同时继续处理别的工作，最后汇总各方结果。
适合的场景
- 任务能拆成几件互不依赖的事，例如同时调研若干个库
- 希望得到几份独立的产出，例如每个模块一份小结
- 范围较大的查找，拆开能明显缩短等待
需要注意
每个并行实例都会各自消耗模型额度。它们在各自的范围内工作，结果由主智能体整合后统一交付。
Markdown 原文Raw markdown
Subagents
Let several agent instances work on independent parts at once.
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.