Not Diamond Model Select API

openapi: 3.0.1
info:
  title: Not Diamond API
  description: >-
    REST API for the Not Diamond AI model router. The modelSelect endpoint routes a
    prompt to the best LLM across providers based on quality, cost, and latency
    tradeoffs. Additional endpoints list supported models, report feedback and
    latency metrics to personalize routing in real time, and train custom routers
    from evaluation datasets.
  termsOfService: https://www.notdiamond.ai/terms
  contact:
    name: Not Diamond
    url: https://www.notdiamond.ai
  version: '2.0'
servers:
  - url: https://api.notdiamond.ai/v2
paths:
  /modelRouter/modelSelect:
    post:
      operationId: modelSelect
      tags:
        - Model Routing
      summary: Route a prompt to the best model.
      description: >-
        Given OpenAI-format messages and a list of candidate LLM providers, returns
        the recommended provider/model along with a session ID for the routing
        decision. Supports cost/latency tradeoffs and personalized routing via a
        preference ID.
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/ModelSelectRequest'
      responses:
        '200':
          description: The recommended provider(s) and a session ID.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/ModelSelectResponse'
        '401':
          description: Missing or invalid API key.
        '422':
          description: Validation error.
  /models:
    get:
      operationId: listModels
      tags:
        - Models
      summary: List supported models.
      description: >-
        Lists all supported text generation models with provider, context length,
        and per-million-token input/output pricing. Results are cacheable for one
        hour.
      parameters:
        - name: provider
          in: query
          required: false
          description: Filter by one or more provider names.
          schema:
            type: array
            items:
              type: string
        - name: openrouter_only
          in: query
          required: false
          description: Return only OpenRouter-supported models.
          schema:
            type: boolean
            default: false
      responses:
        '200':
          description: A list of supported models.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/ModelsListResponse'
  /report/metrics/feedback:
    post:
      operationId: reportFeedback
      tags:
        - Feedback
      summary: Report outcome feedback for a routing session.
      description: >-
        Submits binary outcome feedback (0 or 1) for the provider selected in a
        routing session so Not Diamond can personalize future routing.
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/FeedbackRequest'
      responses:
        '200':
          description: Feedback accepted.
        '401':
          description: Missing or invalid API key.
        '422':
          description: Validation error.
  /report/metrics/latency:
    post:
      operationId: reportLatency
      tags:
        - Feedback
      summary: Report latency metrics for a routing session.
      description: >-
        Reports observed latency (for example tokens per second) for the provider
        selected in a routing session.
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/LatencyRequest'
      responses:
        '200':
          description: Latency metrics accepted.
        '401':
          description: Missing or invalid API key.
        '422':
          description: Validation error.
  /pzn/trainCustomRouter:
    post:
      operationId: trainCustomRouter
      tags:
        - Custom Routers
      summary: Train a custom router.
      description: >-
        Trains a custom router from an uploaded CSV evaluation dataset. Training is
        asynchronous and typically takes 5-15 minutes. Returns a preference ID used
        to drive personalized routing in subsequent modelSelect calls.
      requestBody:
        required: true
        content:
          multipart/form-data:
            schema:
              $ref: '#/components/schemas/TrainCustomRouterRequest'
      responses:
        '200':
          description: The trained router's preference ID.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/TrainCustomRouterResponse'
        '401':
          description: Missing or invalid API key.
        '422':
          description: Validation error.
  /preferences/userPreferenceCreate:
    post:
      operationId: createUserPreference
      tags:
        - Custom Routers
      summary: Create a user preference.
      description: >-
        Creates a user preference (router) and returns its preference ID, which can
        be referenced in modelSelect calls and custom router training.
      requestBody:
        required: false
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/UserPreferenceCreateRequest'
      responses:
        '200':
          description: The created preference ID.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/UserPreferenceCreateResponse'
        '401':
          description: Missing or invalid API key.
components:
  securitySchemes:
    bearerAuth:
      type: http
      scheme: bearer
      description: Not Diamond API key passed as a Bearer token.
  schemas:
    Message:
      type: object
      description: An OpenAI-format chat message.
      properties:
        role:
          type: string
          example: user
        content:
          type: string
      required:
        - role
        - content
    LLMProvider:
      type: object
      description: A candidate provider/model that the router may select.
      properties:
        provider:
          type: string
          description: Provider identifier (e.g. openai, anthropic). Omit when using an OpenRouter slug.
          example: openai
        model:
          type: string
          description: Model identifier, or an OpenRouter slug (e.g. openai/gpt-4o).
          example: gpt-4o
        is_custom:
          type: boolean
          description: Set to true to describe a custom model.
        context_length:
          type: integer
          description: Maximum context length (required for custom models).
        input_price:
          type: number
          description: Cost per million input tokens in USD (required for custom models).
        output_price:
          type: number
          description: Cost per million output tokens in USD (required for custom models).
        latency:
          type: number
          description: Expected latency for the custom model.
      required:
        - model
    ModelSelectRequest:
      type: object
      properties:
        messages:
          type: array
          description: Array of OpenAI-format message objects.
          items:
            $ref: '#/components/schemas/Message'
        llm_providers:
          type: array
          description: List of candidate LLM providers to route between. At least one required.
          items:
            $ref: '#/components/schemas/LLMProvider'
        tools:
          type: array
          nullable: true
          description: OpenAI-format function-calling tools.
          items:
            type: object
        preference_id:
          type: string
          nullable: true
          description: Preference ID for personalized or custom routing.
        tradeoff:
          type: string
          nullable: true
          description: Optimization tradeoff strategy.
          enum:
            - cost
            - latency
        cost_quality_tradeoff:
          type: integer
          nullable: true
          minimum: 0
          maximum: 10
          description: Continuous cost/quality blend (0 = quality, 10 = cost).
        hash_content:
          type: boolean
          default: false
          description: Whether to hash message content for privacy.
        metric:
          type: string
          default: accuracy
          description: Optimization metric for model selection.
        max_model_depth:
          type: integer
          nullable: true
          description: Maximum number of models to consider for routing.
        previous_session:
          type: string
          nullable: true
          description: Previous session ID to link related requests.
      required:
        - messages
        - llm_providers
    ModelSelectResponse:
      type: object
      properties:
        providers:
          type: array
          description: List containing the selected provider/model.
          items:
            $ref: '#/components/schemas/LLMProvider'
        session_id:
          type: string
          description: Unique session ID for this routing decision.
    Model:
      type: object
      properties:
        provider:
          type: string
          example: openai
        model:
          type: string
          example: gpt-4o
        context_length:
          type: integer
        input_price:
          type: number
          description: Cost per million input tokens in USD.
        output_price:
          type: number
          description: Cost per million output tokens in USD.
        openrouter_model:
          type: string
          nullable: true
          description: OpenRouter slug if supported.
    ModelsListResponse:
      type: object
      properties:
        models:
          type: array
          items:
            $ref: '#/components/schemas/Model'
        total:
          type: integer
        deprecated_models:
          type: array
          items:
            $ref: '#/components/schemas/Model'
    FeedbackRequest:
      type: object
      properties:
        session_id:
          type: string
          description: The session ID returned by modelSelect.
        provider:
          $ref: '#/components/schemas/LLMProvider'
        feedback:
          type: object
          description: Feedback payload. The outcome value must be 0 or 1.
          properties:
            value:
              type: integer
              enum:
                - 0
                - 1
      required:
        - session_id
        - provider
        - feedback
    LatencyRequest:
      type: object
      properties:
        session_id:
          type: string
        provider:
          $ref: '#/components/schemas/LLMProvider'
        feedback:
          type: object
          properties:
            tokens_per_second:
              type: number
      required:
        - session_id
        - provider
        - feedback
    TrainCustomRouterRequest:
      type: object
      properties:
        dataset_file:
          type: string
          format: binary
          description: CSV file with evaluation data (minimum 25 samples).
        language:
          type: string
          enum:
            - english
            - multilingual
        llm_providers:
          type: string
          description: JSON array of model providers.
        prompt_column:
          type: string
          description: CSV column name containing prompts.
        maximize:
          type: boolean
          description: Whether higher scores indicate better performance.
        preference_id:
          type: string
          nullable: true
          description: Existing router ID to update.
        override:
          type: boolean
          nullable: true
          default: true
          description: Override existing router.
      required:
        - dataset_file
        - language
        - llm_providers
        - prompt_column
        - maximize
    TrainCustomRouterResponse:
      type: object
      properties:
        preference_id:
          type: string
          description: Unique identifier for the trained router.
    UserPreferenceCreateRequest:
      type: object
      properties:
        name:
          type: string
          nullable: true
          description: Optional name for the preference/router.
    UserPreferenceCreateResponse:
      type: object
      properties:
        preference_id:
          type: string
security:
  - bearerAuth: []