Dify's OpenAI-API-compatible plugin, field by field, for any /v1 endpoint

[apigoto]

I work on the APIGOTO team. APIGOTO is an independent OpenAI-compatible gateway, not affiliated with OpenAI, Anthropic or Google. The plugin form is the same for any OpenAI-compatible endpoint.

Check the endpoint first

curl https://api.china-aiapi.com/v1/models -H "Authorization: Bearer $KEY"

Note the exact model IDs it returns.

Add an LLM

1. Integrations → Model Provider. If OpenAI-API-compatible (langgenius) isn't installed, install it from the Marketplace.
2. On its card, click Add Model and pick model type LLM.
3. Model Name: MODEL_ID, exactly as /models returned it. It's sent as model; Model display name is just the Dify label. Leave model name for API endpoint empty unless you need to send a different ID.
4. API Key: your key.
5. API Base URL: https://api.china-aiapi.com/v1. The plugin appends /chat/completions, so don't add it.
6. Completion mode: Chat. Model context size: the model's real window, not a guess.
7. Capabilities, set per model: Function Call Type (Tool Call / Function Call / Not Support), Stream function calling, Vision Support. Agent nodes rely on these, so only enable what the model supports.
8. Save. Dify sends a small validation request before accepting the credentials.

Repeat 2–8 for each model ID.

Troubleshooting

• Validation fails with 404: almost always the URL. /v1 must appear exactly once; a missing /v1 or a pasted /chat/completions both break it.
• Embeddings / rerank / TTS return 404: the plugin README says these model types add the API version themselves, so their base URL should leave out /v1 to avoid /v1/v1.
• "Model not found": Model Name doesn't match an ID from /models (case and suffix count). The display name is never sent.
• Agent node errors or loops on tools: try the other Function Call Type, or turn off Stream function calling. Tool-call streaming varies between model families even behind one endpoint.
• Errors about max tokens: set Token parameter name to max_tokens or max_completion_tokens instead of Auto.

We use one key for several model families, with upstream failover (at most 3 retries) and a per-request log to see what a workflow sent. Pay-as-you-go, BYOK supported: www.china-aiapi.com

If you run several model families in one workflow, do you keep one provider entry per family or one per use case?