Endpoints
Tianshu offers two routes — just pick one based on the type of request.
Standard route (default)
https://tian-shu.orgSuitable for the vast majority of endpoints: chat, text, embeddings, rerank, and so on. Streaming chat output starts returning tokens as soon as the connection is established and is not affected by first-byte timeouts, so this route is recommended by default.
Direct route (long requests / image generation)
https://api.tian-shu.orgSuitable for long-running requests that return all at once, typically image generation (/v1/images/generations etc.). These requests only return their result after the server finishes computing, which can easily trigger CDN request timeouts; the direct route bypasses the CDN and reaches the origin directly, avoiding timeout interruptions.
Rule of thumb: streaming requests that keep emitting tokens → standard route; requests that compute for a long time and then return all at once (image generation, etc.) → direct route.
Endpoint paths
Both routes are compatible with OpenAI-style paths, for example:
| Use | Path |
|---|---|
| Chat completions | /v1/chat/completions |
| Responses | /v1/responses |
| Claude Messages | /v1/messages |
| Embeddings | /v1/embeddings |
| Rerank | /v1/rerank |
| Image generation | /v1/images/generations |
| Speech / transcription | /v1/audio/speech, /v1/audio/transcriptions |
Which endpoints a specific model supports is determined by what's noted on the Model Marketplace.