[Request] More providers options

[LiteSoul]

First of all, Interesting project, specially that you guys made it easy to customize it.

Ok so I believe it should be easy to open it up to more AI Providers? (like OpenCode already does)
E.g. OpenRouter, Qwen, Nvidia, etc.

Many of use use several providers, thanks!

[SchmidL]

support that fully - my openai compatible key is getting refused -> says API key rejected. Run /auth to update it.

[Simonstj]

Brilliant project but agree... I'll be in when I can use deepseek,Qwen etc subscriptiond

[mehedihassanome]

Please add more providers

[KerrKerrKerr]

At least openrouter, llama.cpp/vllm comatible ones are absolutely needed for those who do not have a kidney to spare on tokens

[nicko-ai]

Implemented, you can test it: npx @vrsen/openswarm@rc

I'll release a new stable version soon!

Additional info:

• more providers are now handled by agentswarm-cli / agency-swarm.
• See releases agentswarm-cli v1.4.37, v1.4.38, plus Ollama PRs #265 and #266.

[lianzhiweivincent-cloud]

Nice to see this moving in the rc.

For DeepSeek/Qwen/OpenAI-compatible providers, I would test the provider support with a small acceptance matrix before the stable release:

1. Custom base_url is preserved exactly and never falls back to api.openai.com after auth refresh.
2. Raw /chat/completions works with a provider-specific model id such as deepseek-chat before it is wrapped by the app.
3. The model id is not normalized into an OpenAI model name internally.
4. Auth errors distinguish invalid API key from unsupported provider shape. A lot of users report custom keys as "rejected" when the real issue is endpoint/model routing.
5. Streaming and non-streaming are tested separately, because some OpenAI-compatible providers pass one path and fail the other.

I keep a provider-agnostic setup checklist here that may help users testing Chinese model endpoints in tools like this: https://black-eagle-ai.vercel.app/integrations/

Disclosure: I work on Black Eagle AI, a Chinese-model gateway. Sharing because the same base_url / model-id problems come up repeatedly when users try DeepSeek or Qwen in OpenAI-compatible tools.