[Bug]: Podcast generation fails with "Provider 'openai_compatible' not supported" error

[EddiePengg]

What did you do when it broke?

Description
When using a custom model provider with the name openai_compatible, podcast generation fails. The system returns an error indicating that only openai-compatible (with hyphen) is supported.

Error Message
ValueError: Provider 'openai_compatible' not supported for language. Supported providers: ['openai', 'openai-compatible', 'anthropic', 'google', 'groq', 'ollama', 'openrouter', 'xai', 'perplexity', 'azure', 'mistral', 'deepseek', 'vertex']

Steps to Reproduce

1. Configure a custom model with provider set to openai_compatible
2. Try to generate a podcast
3. Generation fails with empty outline and transcript

Expected Behavior
Either:

• Both openai_compatible and openai-compatible should be accepted, OR
• The UI should enforce the correct format (openai-compatible) to prevent this error

Environment

• Open Notebook version: v1.7.0
• Deployment: Docker Compose
  Suggestion
  Consider normalizing the provider name (replacing _ with -) or documenting that only hyphen format is valid.
  ▣ Build · glm-5 · 14.7s

How did it break?

null

Logs or Screenshots

Open Notebook Version

v1-latest (Docker)

Environment

No response

Additional Context

No response

Contribution

• I am a developer and would like to work on fixing this issue (pending maintainer approval)

[albkirk]

Hello,

I'm also getting this error.
This may be due to the changes introduced in version 1.7.0, where MODELs (LLMs) have a completely new way of being configured.
It works fine with the previous version 1.6.x. I haven't changed anything from the "llama.ccp" server side.

[newcolour]

+1
seeing the same issue in 1.7.0.

[lfnovo]

Heu guys, I am replacing 1.7.1 image with a fix. If you pull it again in 30 min you should be good to go. Please let me know if it worked?

[EddiePengg]

Heu guys, I am replacing 1.7.1 image with a fix. If you pull it again in 30 min you should be good to go. Please let me know if it worked?

Podcast generation still fails with the following error:

Despite my attempts to resolve the issue by creating a new restart and deleting the previous environment variables, and even after configuring through webui, the problem persists.

DeepSeek Error:
ValueError: DeepSeek API key not found. Set the DEEPSEEK_API_KEY environment variable.

OpenAI Compatible Error:
ValueError: OpenAI-compatible base URL is required. Set OPENAI_COMPATIBLE_BASE_URL_LLM or OPENAI_COMPATIBLE_BASE_URL environment variable or provide base_url in config.

[EddiePengg]

Interestingly, when I click Test in webui, everything seems fine. However, podcast generation always completes immediately. After checking the Docker logs, I found the issue I just mentioned. Due to these problems, I haven't been able to generate a single podcast yet. 😭 Looking forward to your fix.

[albkirk]

Hello!

Thanks a lot for the quick fix!
I'm using Docker Compose to create the Open Notebook instance. As in version 1.7.x we can define the models in the GUI; it feels there's no need to add anything in the compose.yml file.
But, this aproach doesn't work, as EddiePengg showed.
However, if I keep the same configuration as used on v1.6.2:
AI_PROVIDER=openai_compatible
OPENAI_COMPATIBLE_BASE_URL=http://192.168.88.231:8080/v1
OPENAI_COMPATIBLE_API_KEY=
Then, create the models in the webpage (using the same values as in the yaml), and it passes the error above.
A few steps further, it bumps into a TTS-related error.
By adding the line:
OPENAI_COMPATIBLE_BASE_URL_TTS=http://192.168.88.231:8000/v1
I succeeded in creating an audio podcast!

Relevant parameters to keep on v1.7.1:
AI_PROVIDER=openai_compatible
OPENAI_COMPATIBLE_BASE_URL=http://192.168.88.231:8080/v1
OPENAI_COMPATIBLE_API_KEY=
OPENAI_COMPATIBLE_BASE_URL_TTS=http://192.168.88.231:8000/v1

It's a fully functional workaround until you completely fix the issue.