Skip to main content

​
Overview

When logging requests that use structured outputs (JSON schemas), you need to include the schema configuration in the parameters field of your /log-request call. This allows PromptLayer to properly track and display your structured output configurations alongside your request history.

​
When to Use This

Use structured output logging when you’re:
  • Making API calls with JSON schema validation (OpenAI, Anthropic, Google, etc.)
  • Tracking requests that enforce specific response formats
  • Analyzing how different schema configurations affect model outputs
  • Building applications that require reliable, parseable JSON responses

​
Basic Structure

The structured output configuration goes in the parameters field of your log request, using the response_format key with json_schema configuration: 
{
  "provider": "openai",
  "model": "gpt-4",
  "input": {...},
  "output": {...},
  "parameters": {
    "temperature": 0.7,
    "response_format": {
      "type": "json_schema",
      "json_schema": {
        "name": "YourSchemaName",
        "description": "Description of what this schema represents",
        "schema": {
          "type": "object",
          "properties": {
            // Your JSON schema properties
          },
          "required": ["field1", "field2"],
          "additionalProperties": false
        },
        "strict": true
      }
    }
  }
}

​
Complete Examples

​
OpenAI with Structured Outputs

import promptlayer
from datetime import datetime

promptlayer.api_key = "your_api_key"

# Make your OpenAI call with structured outputs
response = openai.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "Generate a recipe for chocolate cake"}],
    response_format={
        "type": "json_schema",
        "json_schema": {
            "name": "Recipe",
            "description": "A structured recipe format",
            "schema": {
                "type": "object",
                "properties": {
                    "recipe_name": {"type": "string"},
                    "ingredients": {
                        "type": "array",
                        "items": {
                            "type": "object",
                            "properties": {
                                "name": {"type": "string"},
                                "amount": {"type": "string"}
                            },
                            "required": ["name", "amount"]
                        }
                    },
                    "instructions": {
                        "type": "array",
                        "items": {"type": "string"}
                    }
                },
                "required": ["recipe_name", "ingredients", "instructions"],
                "additionalProperties": False
            },
            "strict": True
        }
    }
)

# Log the request to PromptLayer
promptlayer.log_request(
    provider="openai",
    model="gpt-4",
    input={
        "type": "chat",
        "messages": [{"role": "user", "content": "Generate a recipe for chocolate cake"}]
    },
    output={
        "type": "chat",
        "messages": [{"role": "assistant", "content": response.choices[0].message.content}]
    },
    request_start_time=datetime.now(),
    request_end_time=datetime.now(),
    parameters={
        "response_format": {
            "type": "json_schema",
            "json_schema": {
                "name": "Recipe",
                "description": "A structured recipe format",
                "schema": {
                    "type": "object",
                    "properties": {
                        "recipe_name": {"type": "string"},
                        "ingredients": {
                            "type": "array",
                            "items": {
                                "type": "object",
                                "properties": {
                                    "name": {"type": "string"},
                                    "amount": {"type": "string"}
                                },
                                "required": ["name", "amount"]
                            }
                        },
                        "instructions": {
                            "type": "array",
                            "items": {"type": "string"}
                        }
                    },
                    "required": ["recipe_name", "ingredients", "instructions"],
                    "additionalProperties": False
                },
                "strict": True
            }
        }
    },
    tags=["structured-output", "recipe-generation"]
)

​
Google Gemini with Structured Outputs

import promptlayer
from datetime import datetime

promptlayer.api_key = "your_api_key"

# Log a Google Gemini request with structured outputs
promptlayer.log_request(
    provider="google",
    model="gemini-2.0-flash-lite-preview-02-05",
    input={
        "type": "chat",
        "messages": [{"role": "user", "content": "Create a product listing"}]
    },
    output={
        "type": "chat",
        "messages": [{"role": "assistant", "content": "..."}]
    },
    request_start_time=datetime.now(),
    request_end_time=datetime.now(),
    parameters={
        "response_format": {
            "type": "json_schema",
            "json_schema": {
                "name": "ProductListing",
                "description": "Product information structure",
                "schema": {
                    "type": "object",
                    "properties": {
                        "product_name": {"type": "string"},
                        "price": {"type": "number"},
                        "description": {"type": "string"},
                        "features": {
                            "type": "array",
                            "items": {"type": "string"}
                        }
                    },
                    "required": ["product_name", "price"],
                    "additionalProperties": False
                },
                "strict": False
            }
        },
        "temperature": 0,
        "maxOutputTokens": 256
    },
    tags=["google", "structured-output"]
)

​
JavaScript/TypeScript Example

import Promptlayer from 'promptlayer';

const promptlayer = new Promptlayer({ apiKey: 'your_api_key' });

// Make your API call with structured outputs
const response = await openai.chat.completions.create({
  model: 'gpt-4',
  messages: [{ role: 'user', content: 'Extract contact information' }],
  response_format: {
    type: 'json_schema',
    json_schema: {
      name: 'ContactInfo',
      description: 'Structured contact information',
      schema: {
        type: 'object',
        properties: {
          name: { type: 'string' },
          email: { type: 'string' },
          phone: { type: 'string' }
        },
        required: ['name', 'email'],
        additionalProperties: false
      },
      strict: true
    }
  }
});

// Log to PromptLayer
await promptlayer.logRequest({
  provider: 'openai',
  model: 'gpt-4',
  input: {
    type: 'chat',
    messages: [{ role: 'user', content: 'Extract contact information' }]
  },
  output: {
    type: 'chat',
    messages: [{ role: 'assistant', content: response.choices[0].message.content }]
  },
  request_start_time: new Date().toISOString(),
  request_end_time: new Date().toISOString(),
  parameters: {
    response_format: {
      type: 'json_schema',
      json_schema: {
        name: 'ContactInfo',
        description: 'Structured contact information',
        schema: {
          type: 'object',
          properties: {
            name: { type: 'string' },
            email: { type: 'string' },
            phone: { type: 'string' }
          },
          required: ['name', 'email'],
          additionalProperties: false
        },
        strict: true
      }
    }
  },
  tags: ['structured-output', 'contact-extraction']
});

​
REST API Example

If you’re calling the /log-request endpoint directly via REST API: 
curl -X POST https://api.promptlayer.com/log-request \
  -H "Content-Type: application/json" \
  -H "X-API-KEY: your_api_key" \
  -d '{
    "provider": "openai",
    "model": "gpt-4",
    "input": {
      "type": "chat",
      "messages": [{"role": "user", "content": "Generate a user profile"}]
    },
    "output": {
      "type": "chat",
      "messages": [{"role": "assistant", "content": "{\"name\":\"John\",\"age\":30}"}]
    },
    "request_start_time": "2024-01-15T10:00:00Z",
    "request_end_time": "2024-01-15T10:00:02Z",
    "parameters": {
      "temperature": 0.7,
      "response_format": {
        "type": "json_schema",
        "json_schema": {
          "name": "UserProfile",
          "description": "User profile information",
          "schema": {
            "type": "object",
            "properties": {
              "name": {"type": "string"},
              "age": {"type": "number"},
              "email": {"type": "string"}
            },
            "required": ["name", "age"],
            "additionalProperties": false
          },
          "strict": true
        }
      }
    },
    "tags": ["user-profile", "structured-output"]
  }'

​
Schema Configuration Details

​
Key Fields

  • name (required): A descriptive name for your schema (e.g., “Recipe”, “ContactInfo”)
  • description (optional): Explains what the schema represents
  • schema (required): The actual JSON schema definition following JSON Schema specification
  • strict (optional): When true, enforces strict validation (supported by some providers like OpenAI)

​
Schema Best Practices

  1. Use clear property names: Make your schema self-documenting
  2. Specify required fields: Use the required array to mark mandatory properties
  3. Set additionalProperties: false: Prevents unexpected fields in the response
  4. Use descriptive types: Leverage JSON Schema’s type system (string, number, array, object, boolean)
  5. Add validation constraints: Use minLength, maxLength, minimum, maximum, etc. when appropriate

​
Tracking Schema Variations

Use tags and metadata to track different schema versions: 
promptlayer.log_request(
    # ... other parameters ...
    parameters={
        "response_format": {
            "type": "json_schema",
            "json_schema": {
                "name": "Recipe_v2",  # Version in name
                # ... schema definition ...
            }
        }
    },
    tags=["structured-output", "recipe-v2", "production"],
    metadata={
        "schema_version": "2.0",
        "environment": "production"
    }
)

​
Common Issues and Solutions

​
Issue: Schema not showing in PromptLayer dashboard

Solution: Ensure the response_format is nested correctly within the parameters field, not at the top level of your log request.

​
Issue: Provider rejects the schema

Solution: Different providers have different schema support levels:
  • OpenAI: Full support with strict: true mode
  • Anthropic: Supports basic JSON mode
  • Google: Supports schemas with some limitations
Check your provider’s documentation for specific schema requirements.

​
Issue: Response doesn’t match schema

Solution:
  1. Verify your schema is valid JSON Schema
  2. Test with strict: true if your provider supports it
  3. Check that your prompt clearly instructs the model about the expected format
  4. Review logged requests in PromptLayer to debug schema mismatches

​
See Also