API overview

This reference documents common text and multimodal endpoints using one standard OpenAPI 3.0 file. Shared field descriptions are bilingual to keep both languages on the same schema.

Endpoint	Purpose
GET /v1/models	List models visible to the current key
POST /v1/chat/completions	Generate a reply from a message array
POST /v1/responses	Use Responses input, output, and event structures
POST /v1/messages	Native Claude messages
`POST /v1beta/models/{model}:generateContent`	Native Gemini generation
`POST /v1beta/models/{model}:streamGenerateContent`	Native Gemini SSE
POST /v1/images/generations	Generate images
POST /v1/images/edits	Edit images
POST /v1/audio/speech	Generate speech
POST /v1/audio/transcriptions	Transcribe audio
POST /v1/audio/translations	Translate audio
POST /v1/videos	Create video task
`GET /v1/videos/{task_id}`	Retrieve video task
`GET /v1/videos/{task_id}/content`	Download video
`HEAD /v1/videos/{task_id}/content`	Inspect video content
`POST /v1/videos/{video_id}/remix`	Remix video
POST /v1/embeddings	Generate text vectors
POST /v1/rerank	Rerank candidate documents
`GET /v1/models/{model}`	Retrieve one model
POST /v1/responses/compact	Compact context
`GET /v1/responses/{response_id}`	Retrieve a plugin task response
POST /v1/moderations	Moderate content
POST /v1/completions	Create a text completion
`POST /v1beta/models/{model}:embedContent`	Gemini single embedding
`POST /v1beta/models/{model}:batchEmbedContents`	Gemini synchronous batch embeddings
`POST /v1/tasks/{key}`	Submit a generic plugin task
`GET /v1/tasks/{key}`	Retrieve a task
`GET /v1/tasks/{key}/artifacts`	List task artifacts
`GET /v1/tasks/{key}/artifacts/{artifact_key}/content`	Download task artifact
`HEAD /v1/tasks/{key}/artifacts/{artifact_key}/content`	Inspect task artifact
POST /v1/alpha/search	Alpha Search request
`GET /mj/task/{id}/fetch`	Retrieve a Midjourney task

WebSocket APIs

Handshake and event references are maintained separately from the 32 REST operations above: Responses WebSocket and Realtime. Find runnable scripts and the shared module in WebSocket integration.

Request conventions

Send Authorization: Bearer YOUR_API_KEY and Content-Type: application/json for JSON requests. For file uploads, use the endpoint’s multipart format and let the client set the boundary. SDK base URLs include /v1. OpenAPI servers contain only the gateway origin and optional prefix because specification paths already include /v1. gateway.example.com is a placeholder. The local reference disables the online request panel. Configure your own server-side client using the integration guide. Native Claude also accepts x-api-key; Gemini accepts x-goog-api-key. Both carry gateway tokens. Native examples use API_ROOT_URL without a protocol prefix; Gemini paths contain /v1beta.

Scope

This curated specification was checked against the implementation. It is not a complete mirror of management, media, or provider-specific endpoints. A listed field does not guarantee unchanged forwarding by every channel or support by every model. Confirm ranges, defaults, tools, and multimodal capabilities for your deployment. Do not mix Chat Completions and Responses fields. Responses may include provider extensions not expanded here; clients should tolerate extra fields. Example outputs illustrate structure only; actual text and usage vary.

Source

Checked revision: 1b7b8df6a2eb3d59485e71ea79a48f248c11b668. Based on project routes, the model-list controller, and relaykit request/response DTOs. Source files and revision are recorded in x-source-files and x-source-revision in the specification. To migrate platforms, reuse the OpenAPI file and replace page openapi frontmatter with the target platform’s API rendering configuration. Errors and retries apply to these endpoints. The added Claude / Gemini operations were checked against d851a5faa49a00abaafd27bf91b311506fbfbc52. Each operation records its own x-source-revision and x-source-files; the original 15 operations retain the baseline above. Runnable native SSE and tool loops for Claude and Gemini include completion validation and signature preservation. Model retrieval, moderation, compaction, and plugin task retrieval were checked at 30a6d88a4, with sources recorded per operation. Plugin task retrieval does not proxy arbitrary upstream responses; see API coverage. 简体中文

Getting started

Integration guides

Application guides

Multimodal guides

Retrieval and reranking

Production and troubleshooting

Additional protocols and tasks

API reference

API overview

WebSocket APIs

Request conventions

Scope

Source

​WebSocket APIs

​Request conventions

​Scope

​Source

WebSocket APIs

Request conventions

Scope

Source