API Reference
Find AGTCloud APIs by development task, then review endpoints, parameters, responses, and interactive testing.
The AGTCloud API Reference documents request URLs, authentication, parameter schemas, response structures, code examples, and interactive testing for each API. Start by finding the API for your task, then open its endpoint page for the complete field definitions.
Making your first API call
If you do not have an API key or have not selected a Base URL and model ID, start with the Quickstart. If your basic setup is complete, choose the API you need below.
Core capabilities
Text, image, and video are the primary API Reference entry points. A capability may offer an OpenAI-compatible format, a provider-specific format, or the AGTCloud asynchronous task format. Choose the format that matches your existing SDK and target model.
Text and chat
Generate text and handle multi-turn conversations with protocols such as Chat Completions, Responses, Anthropic Messages, and Gemini Generate Content.
Image generation
Create, edit, or generate variations of images, and use the Grok Image and Midjourney asynchronous task APIs.
Video generation
Create and retrieve videos through Platform Video Generations, OpenAI Videos, and providers such as Grok, Kling, and Vidu.
More APIs
Audio
Synthesize speech, transcribe and translate audio, run asynchronous speech recognition, and work with voices.
Embeddings
Convert text or token inputs into vectors for semantic retrieval, clustering, and similarity calculations.
Rerank
Rerank candidate documents by relevance with the Jina, Cohere, or Xinference format.
Realtime
Use WebSocket events for realtime text, audio, and function calling.
Choose by API style
Capabilities use different transport methods and result retrieval patterns. Handle data according to the Content-Type and response type documented on each endpoint page.
| API style | Common use cases | Client handling |
|---|---|---|
| Synchronous JSON | Standard text, Embeddings, Rerank | Parse the JSON response when the request completes |
| SSE streaming | Chat, Responses, Gemini streaming | Read events continuously until the stream ends |
| Asynchronous task | Video, Midjourney | Save the task ID, then retrieve its status and result |
| Multipart upload | Image editing, audio transcription | Upload files and form fields as required by the endpoint |
| Binary response | Video or audio content download | Save the byte stream according to its response media type |
| WebSocket | Realtime | Maintain a persistent connection and exchange events continuously |
How to read an endpoint page
Confirm the endpoint and authentication
Check the method, path, API address, and authentication headers required by the endpoint. See Conventions for shared rules.
Choose a request scenario
Select the request example closest to your task, then adjust the Request Body, Path Parameters, or Query Parameters.
Check responses and errors
Parse the result according to its HTTP status and response media type. If a request fails, retain the redacted error details and any Request ID in the response, then see Errors and troubleshooting.
Validate with Try It
Interactive testing uses the API address shown on the current page. Enter an API key for your account and select an available model. Do not use a long-lived production key on a shared device.
Shared rules before integration
- If you are unsure which text protocol to use, read Choose a protocol.
- Replace
MODEL_IDin Reference examples with a model ID available to the current API key. - The same capability may support different fields across models and protocols. Follow the current endpoint, account permissions, and actual response.
- Production applications must not hard-code API keys in browser code or client packages. Try It is for temporary testing only; never place a key in a public repository, screenshot, or log.
Continue reading: Conventions · API keys, Base URL, and authentication