promptfoo--promptfoo/examples/openai-structured-output/promptfooconfig.chat.yaml

# yaml-language-server: $schema=https://promptfoo.dev/config-schema.json
description: 'OpenAI Chat API with External and Inline Schema Examples'
prompts:
  - 'Analyze the following customer support query: "{{query}}" and respond with structured JSON data.'

providers:
  # GPT-5 models with structured output
  - id: openai:chat:gpt-5.4-nano
    label: GPT-5.4 Nano External Schema
    config:
      max_completion_tokens: 800
      seed: 123456
      response_format: file://schema.chat.json # External schema file reference

  - id: openai:chat:gpt-5.4-mini
    label: GPT-5.4 Mini Inline Schema
    config:
      max_completion_tokens: 1000
      response_format:
        type: json_schema
        json_schema:
          name: enhanced_customer_support_analysis
          strict: true
          schema:
            type: object
            properties:
              query_summary:
                type: string
                description: "A brief summary of the customer's query"
              category:
                type: string
                enum:
                  [
                    'billing',
                    'technical_issue',
                    'product_inquiry',
                    'complaint',
                    'feature_request',
                    'feedback',
                    'other',
                  ]
                description: "The main category of the customer's query"
              sentiment:
                type: string
                enum: ['very_positive', 'positive', 'neutral', 'negative', 'very_negative']
                description: "The overall sentiment of the customer's query"
              urgency:
                type: string
                enum: ['1', '2', '3', '4', '5']
                description: 'The urgency level of the query, where 1 is lowest and 5 is highest'
              suggested_actions:
                type: array
                items:
                  type: object
                  properties:
                    action:
                      type: string
                      description: 'A specific action to be taken'
                    priority:
                      type: string
                      enum: ['low', 'medium', 'high', 'critical']
                    estimated_time:
                      type: string
                      description: 'Estimated time to complete this action'
                  required: ['action', 'priority', 'estimated_time']
                  additionalProperties: false
              estimated_resolution_time:
                type: string
                description: "Estimated time to resolve the query (e.g., '2 hours', '1 day')"
              follow_up_required:
                type: boolean
                description: 'Whether a follow-up is required after initial resolution'
            required:
              [
                'query_summary',
                'category',
                'sentiment',
                'urgency',
                'suggested_actions',
                'estimated_resolution_time',
                'follow_up_required',
              ]
            additionalProperties: false

  # Legacy models for comparison
  - id: openai:chat:gpt-4.1-nano
    label: GPT-4.1 Nano External Schema
    config:
      temperature: 0.2
      max_tokens: 800
      seed: 123456
      response_format: file://schema.chat.json # External schema file reference

  - id: openai:chat:gpt-4.1-mini
    label: GPT-4.1 Mini Inline Schema
    config:
      temperature: 0.3
      max_tokens: 1000
      response_format:
        type: json_schema
        json_schema:
          name: enhanced_customer_support_analysis
          strict: true
          schema:
            type: object
            properties:
              query_summary:
                type: string
                description: "A brief summary of the customer's query"
              category:
                type: string
                enum:
                  [
                    'billing',
                    'technical_issue',
                    'product_inquiry',
                    'complaint',
                    'feature_request',
                    'feedback',
                    'other',
                  ]
                description: "The main category of the customer's query"
              sentiment:
                type: string
                enum: ['very_positive', 'positive', 'neutral', 'negative', 'very_negative']
                description: "The overall sentiment of the customer's query"
              urgency:
                type: string
                enum: ['1', '2', '3', '4', '5']
                description: 'The urgency level of the query, where 1 is lowest and 5 is highest'
              suggested_actions:
                type: array
                items:
                  type: object
                  properties:
                    action:
                      type: string
                      description: 'A specific action to be taken'
                    priority:
                      type: string
                      enum: ['low', 'medium', 'high', 'critical']
                    estimated_time:
                      type: string
                      description: 'Estimated time to complete this action'
                  required: ['action', 'priority', 'estimated_time']
                  additionalProperties: false
              estimated_resolution_time:
                type: string
                description: "Estimated time to resolve the query (e.g., '2 hours', '1 day')"
              follow_up_required:
                type: boolean
                description: 'Whether a follow-up is required after initial resolution'
            required:
              [
                'query_summary',
                'category',
                'sentiment',
                'urgency',
                'suggested_actions',
                'estimated_resolution_time',
                'follow_up_required',
              ]
            additionalProperties: false

tests:
  - vars:
      query: "I've been charged twice for my subscription this month. Can you please refund the extra charge?"

  - vars:
      query: "How do I change my password? I can't find the option in my account settings."