Novita AI Streaming & Webhook API

asyncapi: '2.6.0'
id: 'urn:novita:ai:streaming-and-webhooks'
info:
  title: Novita AI Streaming & Webhook API
  version: '1.0.0'
  description: |
    AsyncAPI 2.6 description of the asynchronous surfaces of the Novita AI platform:

    1. **Server-Sent Events (SSE) streaming** for OpenAI-compatible chat completions
       (`POST /openai/v1/chat/completions` with `stream: true`). Modeled as a
       publish-from-server channel that emits incremental chat completion chunk
       events terminated by the `[DONE]` sentinel.
    2. **Outbound webhook callbacks** for the asynchronous image, video, image
       editing, and upscale tasks created against the `/v3/async/*` and
       `/v3beta/*` task endpoints. When a request includes
       `extra.webhook.url`, Novita AI delivers the completed task result to the
       caller-hosted endpoint with the same payload shape returned by the
       `GET /v3/async/task-result` polling endpoint.

    Scope is deliberately limited to event-driven surfaces. Synchronous REST
    operations (LLM non-streaming chat, embeddings, reranking, batch, GPU
    instance management, Seedream 4.0 synchronous image generation, etc.) are
    described separately in the OpenAPI document referenced from `apis.yml`.
  contact:
    name: Novita AI
    url: https://novita.ai/docs/
  license:
    name: Proprietary
    url: https://novita.ai/legal/terms-of-service
  termsOfService: https://novita.ai/legal/terms-of-service
  tags:
    - name: AI
    - name: LLM
    - name: SSE
    - name: Streaming
    - name: Webhooks
    - name: AsyncTasks
    - name: ImageGeneration
    - name: VideoGeneration
defaultContentType: application/json
servers:
  production:
    url: api.novita.ai
    protocol: https
    description: |
      Novita AI production API host. SSE chat completion streams originate from
      `https://api.novita.ai/openai/v1/chat/completions`. Async task submissions
      target `https://api.novita.ai/v3/async/*` (or `https://api.novita.ai/v3beta/*`
      for select FLUX endpoints), and webhook callbacks are dispatched outbound
      from this host to the URL supplied in `extra.webhook.url`.
    security:
      - bearerAuth: []
  webhook-receiver:
    url: '{webhookHost}'
    protocol: https
    description: |
      Customer-hosted HTTPS endpoint that receives outbound POST callbacks from
      Novita AI when an async task completes. The exact URL is the
      `extra.webhook.url` value supplied at task submission time.
    variables:
      webhookHost:
        description: Customer-controlled hostname registered via `extra.webhook.url`.
        default: webhooks.example.com
channels:
  /openai/v1/chat/completions:
    description: |
      Server-Sent Events stream for OpenAI-compatible chat completions. Opened
      by `POST https://api.novita.ai/openai/v1/chat/completions` with
      `Accept: text/event-stream` and request body `stream: true`. Each `data:`
      frame carries a JSON `chat.completion.chunk` object; the stream terminates
      with the literal frame `data: [DONE]`.
    bindings:
      http:
        type: request
        method: POST
        bindingVersion: '0.3.0'
    subscribe:
      operationId: subscribeChatCompletionStream
      summary: Receive incremental chat completion chunks over SSE.
      description: |
        While the underlying transport is an HTTP response with
        `Content-Type: text/event-stream`, from the client's perspective the
        server publishes a sequence of `chat.completion.chunk` events followed
        by a `[DONE]` sentinel.
      tags:
        - name: LLM
        - name: SSE
        - name: Streaming
      message:
        oneOf:
          - $ref: '#/components/messages/ChatCompletionChunk'
          - $ref: '#/components/messages/ChatCompletionStreamDone'
  webhook/task-result:
    description: |
      Outbound webhook delivered to the URL provided in `extra.webhook.url` at
      async task submission time. The POST body matches the synchronous
      `GET /v3/async/task-result?task_id={task_id}` response and is dispatched
      once on task completion. Applies to image generation
      (`/v3/async/txt2img`, `/v3/async/img2img`, FLUX, Seedream 3.0,
      Qwen image), image editing (`/v3/async/upscale`,
      `/v3/async/remove-background`, `/v3/async/replace-background`,
      `/v3/async/inpainting`), and video generation
      (`/v3/async/txt2video`, `/v3/async/img2video`,
      `/v3/async/hunyuan-video-fast`, `/v3/async/kling-v2.1-t2v-master`,
      `/v3/async/minimax-hailuo-02`).
    bindings:
      http:
        type: request
        method: POST
        bindingVersion: '0.3.0'
    publish:
      operationId: publishTaskResultWebhook
      summary: Receive completed async task results.
      description: |
        Novita AI POSTs the completed task payload to the customer webhook
        endpoint. The customer endpoint is expected to respond `2xx` to
        acknowledge receipt.
      tags:
        - name: Webhooks
        - name: AsyncTasks
      message:
        $ref: '#/components/messages/TaskResultWebhook'
components:
  securitySchemes:
    bearerAuth:
      type: http
      scheme: bearer
      bearerFormat: API Key
      description: |
        Novita API key passed as `Authorization: Bearer <NOVITA_API_KEY>` on the
        request that opens the SSE stream. Webhook callbacks are outbound from
        Novita AI and are authenticated by the secrecy of the receiver URL.
  messages:
    ChatCompletionChunk:
      name: chatCompletionChunk
      title: Chat Completion Chunk
      summary: A single incremental token/delta frame in an SSE chat completion stream.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/ChatCompletionChunk'
      examples:
        - name: firstChunkRoleDelta
          summary: First frame in a stream, declaring assistant role.
          payload:
            id: chatcmpl-9pZxYabc
            object: chat.completion.chunk
            created: 1748563200
            model: meta-llama/llama-3.1-8b-instruct
            choices:
              - index: 0
                delta:
                  role: assistant
                  content: ''
                finish_reason: null
        - name: contentChunk
          summary: Mid-stream token delivery.
          payload:
            id: chatcmpl-9pZxYabc
            object: chat.completion.chunk
            created: 1748563200
            model: meta-llama/llama-3.1-8b-instruct
            choices:
              - index: 0
                delta:
                  content: 'Hello'
                finish_reason: null
        - name: finalChunkWithUsage
          summary: |
            Terminal chunk emitted when `stream_options.include_usage` is true.
          payload:
            id: chatcmpl-9pZxYabc
            object: chat.completion.chunk
            created: 1748563200
            model: meta-llama/llama-3.1-8b-instruct
            choices:
              - index: 0
                delta: {}
                finish_reason: stop
            usage:
              prompt_tokens: 12
              completion_tokens: 24
              total_tokens: 36
    ChatCompletionStreamDone:
      name: chatCompletionStreamDone
      title: SSE Stream Terminator
      summary: |
        Sentinel frame written by the server as the literal SSE line
        `data: [DONE]` to signal end of stream.
      contentType: text/plain
      payload:
        type: string
        const: '[DONE]'
        description: Literal `[DONE]` terminator frame.
    TaskResultWebhook:
      name: taskResultWebhook
      title: Async Task Result Webhook
      summary: |
        Final task payload delivered to `extra.webhook.url` when an async
        image/video/edit task reaches a terminal status.
      contentType: application/json
      headers:
        type: object
        properties:
          Content-Type:
            type: string
            const: application/json
      payload:
        $ref: '#/components/schemas/TaskResult'
      examples:
        - name: succeededImageTask
          summary: Successful txt2img / upscale-style result.
          payload:
            task:
              task_id: 71dd988e-632e-4339-b217-74bcbe6db0ee
              task_type: TXT_TO_IMG
              status: TASK_STATUS_SUCCEED
              reason: ''
              eta: 0
              progress_percent: 100
            images:
              - image_url: https://faas-output-image.s3.amazonaws.com/.../0.png
                image_url_ttl: '604800'
                image_type: png
                nsfw_detection_result: null
        - name: succeededVideoTask
          summary: Successful video generation result (hunyuan/kling/hailuo).
          payload:
            task:
              task_id: 0e0a6b4c-1f0a-4e2a-9a4a-5c6e7f8a9b0c
              task_type: TXT_TO_VIDEO
              status: TASK_STATUS_SUCCEED
              reason: ''
              eta: 0
              progress_percent: 100
            videos:
              - video_url: https://faas-output-image.s3.amazonaws.com/.../video.mp4
                video_url_ttl: '604800'
                video_type: mp4
        - name: failedTask
          summary: Failed task callback.
          payload:
            task:
              task_id: 71dd988e-632e-4339-b217-74bcbe6db0ee
              task_type: TXT_TO_IMG
              status: TASK_STATUS_FAILED
              reason: model load failed
              eta: 0
              progress_percent: 0
        - name: testModeCallback
          summary: |
            Test-mode callback triggered when `extra.webhook.test_mode.enabled`
            is true. Status mirrors `extra.webhook.test_mode.return_task_status`.
          payload:
            task:
              task_id: test-task
              task_type: TXT_TO_IMG
              status: TASK_STATUS_SUCCEED
              reason: ''
              eta: 0
              progress_percent: 100
  schemas:
    ChatCompletionChunk:
      type: object
      description: |
        OpenAI-compatible streaming chunk. Mirrors the shape returned by
        `POST /openai/v1/chat/completions` when `stream: true`.
      required:
        - id
        - object
        - created
        - model
        - choices
      properties:
        id:
          type: string
          description: Stable identifier shared by every chunk in a single stream.
        object:
          type: string
          const: chat.completion.chunk
        created:
          type: integer
          format: int64
          description: Unix epoch seconds at which the stream was created.
        model:
          type: string
          description: Model identifier that produced the chunk.
        choices:
          type: array
          items:
            $ref: '#/components/schemas/ChatCompletionChunkChoice'
        usage:
          $ref: '#/components/schemas/Usage'
          description: |
            Present only on the final chunk when the request set
            `stream_options.include_usage: true`. Null on intermediate chunks.
    ChatCompletionChunkChoice:
      type: object
      required:
        - index
        - delta
      properties:
        index:
          type: integer
          description: Choice index (always 0 for single-choice requests).
        delta:
          $ref: '#/components/schemas/ChatCompletionDelta'
        finish_reason:
          type: string
          nullable: true
          enum:
            - stop
            - length
            - tool_calls
            - content_filter
            - null
          description: |
            Null until the model finishes; populated on the terminal content
            chunk for the choice.
        logprobs:
          type: object
          nullable: true
          description: 'Per-token log probabilities when `logprobs: true` was set.'
    ChatCompletionDelta:
      type: object
      description: Incremental delta applied to the assistant message.
      properties:
        role:
          type: string
          enum:
            - assistant
          description: Present on the first chunk only.
        content:
          type: string
          description: Token text appended to the assistant message.
        tool_calls:
          type: array
          description: |
            Streaming tool-call deltas emitted when the model invokes a tool
            declared via the `tools` request field.
          items:
            $ref: '#/components/schemas/ToolCallDelta'
    ToolCallDelta:
      type: object
      properties:
        index:
          type: integer
        id:
          type: string
        type:
          type: string
          enum:
            - function
        function:
          type: object
          properties:
            name:
              type: string
            arguments:
              type: string
              description: JSON-encoded argument fragment.
    Usage:
      type: object
      properties:
        prompt_tokens:
          type: integer
        completion_tokens:
          type: integer
        total_tokens:
          type: integer
    TaskResult:
      type: object
      description: |
        Payload shape shared by the `GET /v3/async/task-result` poll response
        and the outbound webhook callback.
      required:
        - task
      properties:
        task:
          $ref: '#/components/schemas/Task'
        images:
          type: array
          description: Present for image generation and image editing tasks.
          items:
            $ref: '#/components/schemas/TaskImage'
        videos:
          type: array
          description: Present for video generation tasks.
          items:
            $ref: '#/components/schemas/TaskVideo'
    Task:
      type: object
      required:
        - task_id
        - status
      properties:
        task_id:
          type: string
          description: Identifier returned by the original async submission.
        task_type:
          type: string
          description: |
            Task category, e.g. `TXT_TO_IMG`, `IMG_TO_IMG`, `UPSCALE`,
            `REMOVE_BACKGROUND`, `REPLACE_BACKGROUND`, `INPAINTING`,
            `TXT_TO_VIDEO`, `IMG_TO_VIDEO`.
        status:
          type: string
          enum:
            - TASK_STATUS_QUEUED
            - TASK_STATUS_PROCESSING
            - TASK_STATUS_SUCCEED
            - TASK_STATUS_FAILED
        reason:
          type: string
          description: Failure detail when `status` is `TASK_STATUS_FAILED`.
        eta:
          type: integer
          description: Estimated seconds remaining.
        progress_percent:
          type: integer
          minimum: 0
          maximum: 100
    TaskImage:
      type: object
      properties:
        image_url:
          type: string
          format: uri
        image_url_ttl:
          type: string
          description: Time-to-live (seconds) for the signed `image_url`.
        image_type:
          type: string
          enum:
            - png
            - webp
            - jpeg
        nsfw_detection_result:
          type: object
          nullable: true
          description: |
            Populated when NSFW detection was enabled on the original task.
    TaskVideo:
      type: object
      properties:
        video_url:
          type: string
          format: uri
        video_url_ttl:
          type: string
        video_type:
          type: string
          enum:
            - mp4
    WebhookConfig:
      type: object
      description: |
        Shape of the `extra.webhook` object supplied on async task submission.
        Documented here for reference; not transmitted on the callback itself.
      properties:
        url:
          type: string
          format: uri
          description: HTTPS endpoint that will receive the task result callback.
        test_mode:
          type: object
          properties:
            enabled:
              type: boolean
            return_task_status:
              type: string
              enum:
                - TASK_STATUS_SUCCEED
                - TASK_STATUS_FAILED