Please provide the complete log.

ps: I guess it may be a docker network configuration problem.

一样的问题

这就是全部日志
E 2025/03/14 18:19:32 openai.go:55: create chat completion failed: error, status code: 400, status: 400 Bad Request, message: Invalid type for 'messages.[3].content': expected one of a string or array of objects, but got an object instead. (request id: 2025031418193220789630037510643)

I have the same issue~

This entry of log is the only one that is related to AI failure. The other logs are unrelated. Please set up a container yourself, it is very easy to reproduce.

actually, just the Aliyun API has this issue, if using Siliconflow API then it will be fine

actually, just the Aliyun API has this issue, if using Siliconflow API then it will be fine

Thanks for the information. It is possible that the Aliyun API is incompatible with the OpenAI API specification in some details. We cannot change this. Thanks for your feedback and discussion.

As stated, it works in windows client, not in docker. This seems like an platform dependent implementation error instead of an error from aliyun from my perspective.

The implementation code of this part has nothing to do with the specific platform and is the same code.

I tried another AI provider: https://api.deepinfra.com/v1/openai
I got the same error in docker container again.

Not sure why this ticket is closed. Could you verify it before closing it? This should be very easy to verify. Just put in any provider in a docker container frontend. You will see the error.

I tested https://api.deepseek.com/v1 with SiYuan Docker and it is available. The configuration is as follows:

I think you could try changing this https://api.deepinfra.com/v1/openai to https://api.deepinfra.com/v1

我这不行，docker依然报错，但桌面端不报错

超时时间改成600试试

超时时间改成600试试

不行，一样的错
E 2025/03/21 17:01:55 openai.go:55: create chat completion failed: error, status code: 422, status: 422 Unprocessable Entity, message: invalid character 'F' looking for beginning of value, body: Failed to deserialize the JSON body into the target type: messages[0]: missing field content at line 1 column 53

你用的是 ai 的哪个功能？

你用的是 ai 的哪个功能？

就问了 “你好”

我这里在 docker 上测试正常：

建议还是检查一下网络相关配置。

终于知道怎么重现了，真不容易

随便填个错误的key，然后用一次，会报错，再把key改回正确的，还是会报错

重启docker后就不报错了

推测应该是缓存问题

只要报一次错，以后无论配置哪家的AI都会报错

重启docker后就不报错了

Thanks for the trick, I rebooted my container by purging it and force recreate, it now works with the https://api.deepinfra.com/v1/openai API. Exact same API, no change in configuration. What always failed now works like nothing happened before.

只要报一次错，以后无论配置哪家的AI都会报错

Really not sure why a failure can persist.

@sz2013 找到问题了，DeepSeek 不支持 content 为空字符串，这个问题下个版本解决，感谢反馈。

@phdbbs Please upgrade to the latest SiYuan