AI continuously retries search after search function errors, consuming large amounts of tokens

[yan2027]

📱 Client Type

Desktop App (Electron)

💻 Operating System

Windows

📦 Deployment Platform

No response

🔧 Deployment Mode

client db (lobe-chat image)

📌 Version

v2.0.0-next.115

🌐 Browser

Edge

🐛 Bug Description

I needed the model to search online for some prices. I didn't enable in-model search because it would produce no output, so I enabled smart networking instead. However, during those ten-plus minutes, it kept searching for gypsum ceiling prices, and after consuming 600,000 tokens I stopped it. Upon checking the search results, I found that errors occurred from the very first result. The model seemed unaware of the problem and continued calling the search tool repeatedly.

LobeHub.2025-11-25.22-48-54.mp4

📷 Recurrence Steps

No response

🚦 Expected Behavior

I hope there can be a setting in the client to configure which search method to use. For example, in the backend settings, you could configure the search service and API. The image shows what I think would be a good approach. Also, if there are issues with search results, the AI model should be able to detect them and stop subsequent calls.

📝 Additional Information

No response

🛠️ Willing to Submit a PR?

No, but I am happy to help test the fix

✅ Validations

• Read the docs.
• Check that there isn't already an issue that reports the same bug to avoid creating a duplicate.
• Make sure this is a LobeChat issue and not a third-party library or provider issue.
• Check that this is a concrete bug. For Q&A, please use GitHub Discussions or join our Discord Server.

---

This issue was translated by Claude.

Original Content

📱 Client Type

Desktop App (Electron)

💻 Operating System

Windows

📦 Deployment Platform

No response

🔧 Deployment Mode

client db (lobe-chat image)

📌 Version

v2.0.0-next.115

🌐 Browser

Edge

🐛 Bug Description

我需要模型联网搜索一些价格,没有开启模型内搜索,因为会无输出,于是开启了智能联网,但是在这十多分钟的时间里,它一直在搜索石膏吊顶的价格,最后消耗六十万tokens后我停止了它,检查搜索结果发现从第一条开始就是报错的,模型似乎没有意识到问题还在持续调用搜索工具

LobeHub.2025-11-25.22-48-54.mp4

📷 Recurrence Steps

No response

🚦 Expected Behavior

希望在客户端有地方可以设置使用何种方法搜索,比如说在后台可以设置搜索服务以及api,图为我觉得挺好的方法,以及如果搜索结果有问题,能让ai模型知道并停止接下来的调用

📝 Additional Information

No response

🛠️ Willing to Submit a PR?

No, but I am happy to help test the fix

✅ Validations

• Read the docs.
• Check that there isn't already an issue that reports the same bug to avoid creating a duplicate.
• Make sure this is a LobeChat issue and not a third-party library or provider issue.
• Check that this is a concrete bug. For Q&A, please use GitHub Discussions or join our Discord Server.

[lobehubbot]

@arvinxx @ONLY-yours - High priority search and tool calling issue. The AI model is continuously retrying failed search tool calls, resulting in massive token consumption (600k tokens). This involves both search functionality and tool calling error handling.

[dosubot]

你的反馈确实反映了当前 LobeChat 搜索工具的一个已知问题：当搜索报错时，AI 可能会持续尝试调用搜索插件，导致大量 token 消耗。这个问题已在 issue #10082 被跟踪，目前还没有正式修复方案，临时解决方法是回退到 v2.0.0-next.16 版本，能避免工具卡死或无限调用的情况。

目前的实现中，搜索插件如果报错，会把错误信息更新到消息上，并停止本轮调用，但如果对话逻辑再次触发搜索，AI 还是可能继续尝试，没有自动的重试/限流机制，也不会自动“意识到”连续失败后彻底停止调用（详见 代码实现）。

你提到的“客户端可配置搜索服务和 API、错误自动终止”等功能，目前还没有在 UI 层实现，所有搜索引擎、API Key、代理等配置只能通过服务端环境变量（如 SEARCH_PROVIDERS、CRAWLER_IMPLS 等）设置，客户端没有类似你截图那样的配置界面（参考 配置代码）。

如果你需要临时规避，可以考虑：

• 回退到 v2.0.0-next.16 版本。
• 通过环境变量配置只保留一个稳定的搜索源，减少出错概率。
• 关注 #10082 和后续版本更新，未来会有更智能的工具调用和错误处理机制（如 PR #10340 提到的“按需调用”架构）。

目前还无法在客户端直接配置搜索服务或设置错误重试/终止策略，但你的建议很有价值，社区也有类似需求在讨论中。

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}