use OpenAI-compatible API, Request Failed 403 Invalid response body

[LuoXingchen935]

App Version

3.23.11,VisualStudioCode Current Release,Ubuntu 24.04.2

API Provider

OpenAI Compatible

Model Used

DeepSeek-R1

Roo Code Task Links (Optional)

No response

🔁 Steps to Reproduce

1.set API,choose OpenAI Compatible
2.enter API URL and API KEY and Model ID
3.Model configuration is default
4.Send "Who are you"
5.API Request failed,403 Invalid response body while trying to fetch https://xxx.xxx.xxx.com/v1/chat/completions: Premature close

💥 Outcome Summary

set OpenAI Compatible, ask "who are you", 403 terminated

📄 Relevant Logs or Errors (Optional)

[dosubot]

A 403 error with "Invalid response body" and "Premature close" usually means the API endpoint refused the request—often due to an authentication or permission issue, or because the endpoint isn't fully compatible with the expected OpenAI API format. For Roo Code, the most common causes are an incorrect Base URL, API Key, or Model ID, or using an endpoint that doesn't fully support the OpenAI-compatible API spec required by Roo Code. Double-check that your Base URL matches the DeepSeek-R1 provider's OpenAI-compatible endpoint, your API Key is valid and has the right permissions, and your Model ID is correct. Also, make sure the provider's API is accessible from your network and that there are no typos in your configuration. If everything looks correct and you still get this error, it's possible the DeepSeek endpoint isn't fully compatible with Roo Code's OpenAI-compatible integration or there may be a provider-side issue. The documentation doesn't mention this specific error, but these are the main troubleshooting steps recommended for connection and compatibility issues (source).

To reply, just mention @dosu.

---

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

[roomote]

I am investigating this OpenAI-compatible API 403 error issue. The problem appears to be related to how Roo Code handles API requests to DeepSeek-R1 endpoints, specifically with "Invalid response body" and "Premature close" errors. I will examine the OpenAI-compatible provider implementation and error handling to identify and fix the root cause.

[daniel-lxs]

Thanks for reporting this. Just to confirm, are you able to access that endpoint directly? A 403 usually means the server is rejecting the request due to permissions or an invalid API key, so it would help to know if you're able to reach https://xxx.xxx.xxx.com/v1/chat/completions from your browser or using something like curl or Postman.

Let us know what you find!

[LuoXingchen935]

@daniel-lxs
Hi,daniel
I used Postman for testing, and the results are shown in the following figure.

The test results show that requests can be made in the chat completions format of OpenAI as normal. And using other front - end dialogue windows of LLM can also get results normally.

The above results all show that the API interface is normal. The conversation in the second picture is the result obtained by having a conversation between two LLM models with context memory turned off.
I have multiple LLM providers. To facilitate management and switching, I use a self - hosted API aggregation tool to forward requests to upstream models.