Packages a Cortex project and deploys each function as a persistent REST endpoint via the Cerebrium CLI (init, login, run, deploy), with apps, deployments, scaling, and configuration managed through the CLI and dashboard.
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: []