[Feature] OpenAI-compatible provider integrations (DeepSeek, Mistral, Groq, Zhipu, MiniMax, Qwen, Moonshot, Doubao)

[JackChen-me]

These providers offer OpenAI-compatible APIs and should work via existing openai provider with custom baseURL. Each one needs:

1. Verify tool-calling works correctly
2. Add a working example to examples/
3. Document in README under Supported Models

Provider	baseURL	Env var
DeepSeek	https://api.deepseek.com	DEEPSEEK_API_KEY
Mistral	https://api.mistral.ai/v1	MISTRAL_API_KEY
Groq	https://api.groq.com/openai/v1	GROQ_API_KEY
Zhipu GLM	https://open.bigmodel.cn/api/paas/v4	ZHIPU_API_KEY
MiniMax	https://api.minimax.chat/v1	MINIMAX_API_KEY
Qwen (DashScope)	https://dashscope.aliyuncs.com/compatible-mode/v1	DASHSCOPE_API_KEY
Moonshot (Kimi)	https://api.moonshot.cn/v1	MOONSHOT_API_KEY
Doubao (ByteDance)	https://ark.cn-beijing.volces.com/api/v3	ARK_API_KEY

Contributions for any single provider are welcome — no need to cover all at once.

[JackChen-me]

How to contribute

Pick a provider, test it, submit a PR. One provider per PR is fine.

Claim a provider

Comment below with the provider name (e.g. "I'll take DeepSeek") to avoid duplicate work.

Provider	Status	Claimed by
DeepSeek	⬜	—
Mistral	⬜	—
Groq	⬜	—
Zhipu GLM	⬜	—
MiniMax	⬜	—
Qwen (DashScope)	⬜	—
Moonshot (Kimi)	⬜	—
Doubao (ByteDance)	⬜	—

What a PR should include

1. Test tool-calling — Run a simple agent with at least one tool (e.g. bash or a custom tool) to confirm the provider handles tool_use / tool_result correctly.

2. Add an example — Create examples/XX-<provider-name>.ts following the pattern in existing examples. Minimal is fine:

   const agent: AgentConfig = {
     name: 'test-agent',
     model: '<model-id>',
     provider: 'openai',
     baseURL: '<provider-base-url>',
     apiKey: process.env.<PROVIDER>_API_KEY,
     systemPrompt: 'You are a helpful assistant.',
     tools: ['bash'],
   }

3. Update README — Add a row to the "Supported Providers" table in both README.md and README_zh.md with status set to Verified.

Reference

See example 06 (local model) for how provider: 'openai' + baseURL works. The same pattern applies to all providers listed here.

Questions? Drop them in this thread.

[namezzb]

Hi! I'd like to work on this issue. I'll verify and document the OpenAI-compatible providers (DeepSeek, Groq, Mistral, Zhipu, MiniMax, Qwen, Moonshot, Doubao).

I'll start by:

1. Testing each provider with the existing adapter
2. Creating a verification report with working configurations
3. Adding documentation to README

Looking forward to contributing!

[tangfeng2]

Hi, could you please add "require": "./dist/index.js" to the exports field in package.json? This will allow the package to work properly with CommonJS projects. Thanks!

[JackChen-me]

Thanks for the suggestion! This package is ESM-only ("type": "module" with ESNext output), so simply adding "require" to exports would cause ERR_REQUIRE_ESM at runtime.

If you are using it in a CommonJS project, dynamic import works:

const { OpenMultiAgent } = await import('open-multi-agent')

No plans to add a dual CJS/ESM build for now -- keeping the build simple. If this becomes a common request we can revisit.

[chocofoxy]

Hello just a suggestion if you are open to consider it, to add more providers support you can wrap the model in the ai/sdk library built by vercel