Skip to content

Baoyu Explains How GrokBot Routes Research Tasks to Threads

Baoyu · X· @dotey·· 1 hours ago

1.2K views · 4 likes · 0 reposts (at collection)

Original
Who
Bao Yu
What
explains how coordinators and threads should divide work across models in GrokBot-style applications
Quote
回到你说的场景在 Grokbot 里面,我在数学研究Bot的主会话说要研究一个数学课题,主会话不会自己开始研究,而是新开一个 Thread,把我的要求都原样转发给 Thread,让 Thread 去做研究。

What happened

Baoyu says a GrokBot main session handles incoming messages and starts a new thread for a specific research task, forwarding the request as written. He compares the main session to a front desk that can use a simpler model, while a stronger model such as Fable plans, schedules and reviews the work, and Opus handles execution.

Read next

See the full storyBaoyu Explains Model Routing and Prompt Caching Across Main Sessions and Threads
Body · English edition pending (original content)

我们两说的场景是不一样的,其实不矛盾

你说的场景更像是一个具体的任务(如果我没理解错),比如我要让 Agent 去研究一个数学课题。

我说的场景是指 GrokBot 这种 Bot 类应用的主会话,就是什么消息你都往一个会话发。

但是实际执行任务的会新建 Thread,把消息转发给 Thread,让 Thread 去处理。

对于这种具体任务,就应该按照你说的,让强的模型去调度,比如 Fable 去计划、调度和验收,Opus 去执行。

回到你说的场景在 Grokbot 里面,我在数学研究Bot的主会话说要研究一个数学课题,主会话不会自己开始研究,而是新开一个 Thread,把我的要求都原样转发给 Thread,让 Thread 去做研究。

这就好比主会话是前台(sonnet),它不需要多懂数学,当它识别到用户是要做一个负责数学任务,它开始一个新的任务(Thread),让数学专家(Fable)去处理,数学专家在执行任务的时候,又去找了几个研究生(Opus)去帮他做一些具体执行的任务。

Quoted postGreen_Orange_GO@gocggo

@dotey 我是用cc做research的,我很好奇官方以及宝玉老师为什么都推荐coordinator用简单一些的模型,thread用强一些的模型。对于我的课题来说数学性较强,我朴素的认为管理这个项目的主agent应该有更强的能力去统筹整个研究并参与决策,而subagent由主agent根据子任务的复杂度分配给不同强度的模型及effort。

View the quoted post on X

The story so far

Baoyu outlined how AI workflows can pair a lightweight main session for task routing with stronger models in separate threads. In Claude Project, he uses Sonnet 5.5 for the main session and Opus 5.5 for subtasks. In a GrokBot math research setup, Baoyu said the main session acts like a front desk that receives incoming messages and opens a new thread for each research request, forwarding the prompt as written. A stronger model such as Fable plans, schedules, and reviews the work, while Opus handles execution. To reduce token consumption, Baoyu recommends sending only recent messages after a main session's prompt cache expires and compressing thread context before its cache expires, such as at 50 minutes for a one-hour cache.