From setup to your first request.
Accounts are invite-only. Browsing is public; inference always requires a valid platform API key.
01 · Prepare credentials and balance
Create a key in the console; its full value is shown once. Available recharge methods are shown on the balance and billing page. Never use an upstream secret in the client.
export TOBASE_API_KEY="<your-platform-key>" # Gateway base URL: <gateway-base-url>
02 · Choose a native protocol
| Protocol | Endpoint |
|---|---|
| Chat Completions | POST /v1/chat/completions |
| Responses | POST /v1/responses |
| Messages | POST /v1/messages |
| Gemini | POST /v1beta/models/{model}:generateContent |
| Gemini SSE | POST /v1beta/models/{model}:streamGenerateContent?alt=sse |
All four accept a Bearer platform key. Messages also accepts x-api-key; Gemini accepts x-goog-api-key. Native parameters and SSE are supported, including tool calls. Available parameters and input types depend on the selected model and upstream API.
Select a supplier using only the provider parameter: in the request body for Chat, Responses and Messages, or in the query for Gemini. Keep the model name unchanged, without a supplier suffix.
03 · Choose a model and copy its example
No published model enables this protocol
04 · Follow usage and charges
Charges use trustworthy input/output usage and request-time rates. Reservations are not spending. Daily charges follow the request registration date in Beijing time (Asia/Shanghai). The platform keeps request IDs, token usage, charges and channel tracing metadata without storing request or response bodies. The model API and Playground return live model responses.
Admission congestion is 503 wallet_admission_busy with Retry-After; 402 requires database-confirmed insufficient balance. Repeated generation may be billed separately. Do not replay a request with an unknown result automatically.