Cerebrium Streaming Endpoints API

openapi: 3.0.1
info:
  title: Cerebrium Cortex Inference API
  description: >-
    Specification of the Cerebrium serverless GPU inference surface. Each
    function deployed with the Cortex framework and the Cerebrium CLI becomes an
    authenticated POST endpoint under
    /v4/{projectId}/{appName}/{functionName}. The same endpoint supports
    synchronous JSON responses, Server-Sent Events streaming (Accept:
    text/event-stream), asynchronous submission via the async=true query
    parameter, and OpenAI-compatible chat/embedding payloads. Endpoint hosts are
    regional (for example api.aws.us-east-1.cerebrium.ai); the legacy
    api.cortex.cerebrium.ai host is also documented. Authentication uses the JWT
    bearer token from the dashboard API Keys section. App deployment, scaling,
    logs, and status are managed through the Cerebrium CLI and dashboard rather
    than a documented public management REST API.
  termsOfService: https://www.cerebrium.ai/terms-of-service
  contact:
    name: Cerebrium Support
    url: https://www.cerebrium.ai/docs
    email: support@cerebrium.ai
  version: 'v4'
servers:
  - url: https://api.aws.us-east-1.cerebrium.ai
    description: AWS us-east-1 regional endpoint (region varies by deployment)
  - url: https://api.cortex.cerebrium.ai
    description: Cortex endpoint host (also documented for v4 invocation)
paths:
  /v4/{projectId}/{appName}/{functionName}:
    post:
      operationId: runFunction
      tags:
        - Inference
      summary: Invoke a deployed function
      description: >-
        Calls a deployed Cortex function. The JSON request body maps directly to
        the function's parameters. Returns a JSON object containing run_id,
        run_time_ms, and result. Use the function name `run` (or `predict`, or
        any public function defined in the app) as documented in examples. To
        stream output, send Accept: text/event-stream and have the function
        yield data; the response is a text/event-stream of `data:` lines.
      parameters:
        - name: projectId
          in: path
          required: true
          description: Cerebrium project identifier, for example p-xxxxxxxx.
          schema:
            type: string
        - name: appName
          in: path
          required: true
          description: Deployed app name from cerebrium.toml.
          schema:
            type: string
        - name: functionName
          in: path
          required: true
          description: >-
            Public function exposed by the app (for example run or predict).
            Functions prefixed with an underscore are private and not callable.
          schema:
            type: string
        - name: async
          in: query
          required: false
          description: >-
            When true, the request is accepted for asynchronous execution and
            the call returns 202 with a run_id instead of the result.
          schema:
            type: boolean
        - name: Accept
          in: header
          required: false
          description: >-
            Set to text/event-stream to receive a Server-Sent Events stream from
            a generator function.
          schema:
            type: string
      requestBody:
        required: false
        content:
          application/json:
            schema:
              type: object
              additionalProperties: true
              description: >-
                Free-form JSON whose keys map to the deployed function's
                parameters.
      responses:
        '200':
          description: Synchronous execution succeeded.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/RunResponse'
            text/event-stream:
              schema:
                type: string
                description: Server-Sent Events stream of `data:` lines.
        '202':
          description: Async request accepted (async=true).
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/AsyncAcceptedResponse'
        '401':
          description: Missing or invalid authentication token.
        '403':
          description: Authenticated token is not authorized for this resource.
        '404':
          description: App or function not found.
        '500':
          description: Application exception or platform error.
  /v4/{projectId}/{appName}/{functionName}/chat/completions:
    post:
      operationId: openaiChatCompletions
      tags:
        - OpenAI Compatible
      summary: OpenAI-compatible chat completions
      description: >-
        OpenAI-compatible chat completions path served by a function that
        accepts the OpenAI request parameters and yields a JSON-serializable
        response. Use the standard OpenAI client with the function base URL and
        the Cerebrium JWT as the api_key.
      parameters:
        - name: projectId
          in: path
          required: true
          schema:
            type: string
        - name: appName
          in: path
          required: true
          schema:
            type: string
        - name: functionName
          in: path
          required: true
          schema:
            type: string
      requestBody:
        required: true
        content:
          application/json:
            schema:
              type: object
              additionalProperties: true
              description: OpenAI-compatible chat completion request body.
      responses:
        '200':
          description: Chat completion response (JSON or streamed).
        '401':
          description: Missing or invalid authentication token.
        '404':
          description: App or function not found.
        '500':
          description: Application exception or platform error.
  /v4/{projectId}/{appName}/{functionName}/embedding:
    post:
      operationId: openaiEmbedding
      tags:
        - OpenAI Compatible
      summary: OpenAI-compatible embeddings
      description: >-
        OpenAI-compatible embeddings path served by a function implementing the
        embeddings interface. Invoked with the standard OpenAI client using the
        function base URL and the Cerebrium JWT.
      parameters:
        - name: projectId
          in: path
          required: true
          schema:
            type: string
        - name: appName
          in: path
          required: true
          schema:
            type: string
        - name: functionName
          in: path
          required: true
          schema:
            type: string
      requestBody:
        required: true
        content:
          application/json:
            schema:
              type: object
              additionalProperties: true
              description: OpenAI-compatible embeddings request body.
      responses:
        '200':
          description: Embedding response.
        '401':
          description: Missing or invalid authentication token.
        '404':
          description: App or function not found.
        '500':
          description: Application exception or platform error.
components:
  securitySchemes:
    bearerAuth:
      type: http
      scheme: bearer
      bearerFormat: JWT
      description: >-
        JWT token from the dashboard API Keys section, sent as
        Authorization: Bearer <JWT_TOKEN>.
  schemas:
    RunResponse:
      type: object
      properties:
        run_id:
          type: string
          description: Unique identifier for the request.
        run_time_ms:
          type: number
          description: Execution duration in milliseconds.
        result:
          description: The data returned by the function.
      required:
        - run_id
        - result
    AsyncAcceptedResponse:
      type: object
      properties:
        run_id:
          type: string
          description: Identifier of the accepted asynchronous run.
      required:
        - run_id
security:
  - bearerAuth: []