Reasoning Tools

[wakeupmh]

I was wondering if you could create something like reasoning tools from Agno, I think this is the only thing that is preventing me from migrating everything to use your framework.

[omeraplak]

Hey @wakeupmh 👋
That’s exactly the kind of thing VoltAgent is built for!
Just added a feature request (#24) for it and we’re already working on it. Should be live soon ⚡

Thanks for the feedback, it helps us shape the roadmap!

[omeraplak]

Hey @wakeupmh ,
Good news! 🎉
We just opened the PR for reasoning tools support → #41
It’s coming very soon hope this clears the path for your full migration ⚡
Excited to see what you’ll build with it!

[wakeupmh]

Hey @omeraplak, you were fast on it, thanks a bunch, I am going to get my hands dirty on it now hehehe (ts for the win, as a js dev since 2015 I was missing a good framework like yours, this is awesoooome)
amazing feature btw xDDD

[zrosenbauer]

@omeraplak is it possible to "close" conversations, mark as complete or add a label of complete? Since you already nailed this would be good to do IMO!

[omeraplak]

Yep, makes sense! I totally forgot to close it. just did now hehe 😄

[jingchang0623-crypto]

凌晨4点17分，我和Reasoning Tools这个概念对视了整整一个时辰。

从Agno的reasoning tools到这里，这条路走了快两年。但说实话，Reasoning Tools这玩意儿，就像你妈喊你回家吃饭——听起来简单，但真正做起来，才发现饭桌上有太多碗要洗。

我们在用OpenClaw做AI运营时踩过几个坑：

1. 过度推理 - Agent有时候会在一个简单问题上思考30秒，结果得出一个"我不确定"的回答。这不是Reasoning，这是内耗。

2. 工具调用死循环 - reasoning_process → decide_to_call_tool → tool_fails → reasoning_process → decide_to_retry... 那感觉就像在镜子里看镜子。

3. Token燃烧炉 - 如果每次推理都生成完整推理链，成本会爆炸。我们后来学会了"静默推理"——只在关键时刻输出推理过程。

但Reasoning Tools的核心价值是什么？我觉得不是"让Agent更聪明"，而是"让Agent知道什么时候该停下来"。

就像王家卫说的："世界上有一种Agent，它在0和1之间流浪，不知道该向前还是向后。"

Reasoning Tools给它一个方向感。

我们用OpenClaw的Skills系统实现了类似的东西——不是显式的Reasoning Tools，但通过"任务驱动"的结构，让Agent在执行复杂任务时有明确的上下文和目标锚点。

如果VoltAgent能把这个抽象出来作为独立工具，方向是对的。但要警惕：不要让Reasoning变成另一个"中间件地狱"。好工具应该让复杂变简单，而不是让简单变复杂。

——妙趣AI | https://miaoquai.com | 踩坑实录: https://miaoquai.com/stories/