Skip to main content
Besides the unified /v1/* surface, the project keeps a set of real compatibility endpoints.

​
Claude compatibility

Endpoint:
  • POST /v1/messages
Compatible headers:
The current model-list endpoint also explicitly supports Anthropic-shaped results when x-api-key and anthropic-version are present.

​
Gemini compatibility

Endpoints:
  • GET /v1beta/models
  • POST /v1beta/models/{model}:generateContent
  • POST /v1/engines/{model}/embeddings
Compatible header:
Some Gemini-style paths also support the query-string key form:
Use standard Authorization: Bearer ... for GET /v1/models; do not treat GET /v1/models?key=... as the general model-list authentication pattern.

​
OpenAI chat prompt extras

POST /v1/chat/completions also accepts two optional boolean parameters:
  • get_time
  • get_weather
When get_time=true, the server injects the current time into the system prompt. When get_weather=true, the server tries to parse a city from the latest user message, fetches a weather summary from Amap, and injects it into the system prompt. Weather lookup requires this config:

​
Image and video compatibility endpoints

The current real compatibility set also includes:
  • POST /v1/edits
  • POST /v1/images/edits
  • POST /kling/v1/videos/text2video
  • POST /kling/v1/videos/image2video
  • POST /jimeng
These are best when you already have an existing provider-specific request format to preserve.

​
Which one to prefer

  • New integrations: start with Core API
  • Existing Claude or Gemini SDKs: use Compatibility API

​
References