Braintrust Evals API

openapi: 3.0.3
info:
  title: Braintrust REST ACL Evals API
  description: The Braintrust REST API for building, evaluating, and observing AI applications. The API is organized around REST, uses predictable resource-oriented URLs under https://api.braintrust.dev/v1, accepts and returns JSON, and authenticates with a Bearer API key. This specification covers the core documented resource surface - projects, experiments, datasets, logs/spans, prompts, functions and scorers, evals, project configuration, organization/ACL management, credentials, and the OpenAI-compatible AI proxy.
  termsOfService: https://www.braintrust.dev/legal/terms-of-service
  contact:
    name: Braintrust Support
    email: support@braintrust.dev
    url: https://www.braintrust.dev/docs
  version: '1.0'
servers:
- url: https://api.braintrust.dev
  description: US data plane (default)
- url: https://api-eu.braintrust.dev
  description: EU data plane
security:
- bearerAuth: []
tags:
- name: Evals
paths:
  /v1/eval:
    post:
      operationId: evalLaunch
      tags:
      - Evals
      summary: Launch an eval
      description: Launch an evaluation. This is the API-equivalent of the Eval function in the SDK; it runs a task function over a dataset and applies scorers.
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/EvalLaunchRequest'
      responses:
        '200':
          description: Returns a summary of the launched eval.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/SummarizeExperimentResponse'
        '401':
          $ref: '#/components/responses/Unauthorized'
components:
  schemas:
    SummarizeExperimentResponse:
      type: object
      properties:
        project_name:
          type: string
        experiment_name:
          type: string
        project_url:
          type: string
        experiment_url:
          type: string
        comparison_experiment_name:
          type: string
          nullable: true
        scores:
          type: object
          nullable: true
          additionalProperties:
            type: object
            properties:
              name:
                type: string
              score:
                type: number
              diff:
                type: number
                nullable: true
              improvements:
                type: integer
              regressions:
                type: integer
        metrics:
          type: object
          nullable: true
          additionalProperties: true
    Error:
      type: object
      properties:
        error:
          type: string
          description: A human-readable error message.
    EvalLaunchRequest:
      type: object
      required:
      - project_id
      - data
      - task
      - scores
      properties:
        project_id:
          type: string
          format: uuid
        experiment_name:
          type: string
        data:
          description: A dataset id, dataset name, or inline dataset definition to evaluate over.
          oneOf:
          - type: object
            additionalProperties: true
          - type: string
        task:
          description: A function id or slug naming the task function to run over each input.
          oneOf:
          - type: object
            additionalProperties: true
          - type: string
        scores:
          type: array
          description: A list of scoring functions (by id, slug, or inline) to apply.
          items:
            oneOf:
            - type: object
              additionalProperties: true
            - type: string
        metadata:
          type: object
          additionalProperties: true
        stream:
          type: boolean
  responses:
    Unauthorized:
      description: Authentication failed - missing or invalid API key.
      content:
        application/json:
          schema:
            $ref: '#/components/schemas/Error'
  securitySchemes:
    bearerAuth:
      type: http
      scheme: bearer
      bearerFormat: API key or JWT
      description: 'Authenticate requests with your Braintrust API key in the Authorization header, e.g. `Authorization: Bearer $BRAINTRUST_API_KEY`.'