Structured Generation

BrowserAI supports generating structured outputs using JSON schemas to enforce specific response formats.

Using JSON Schemas

Define output structure using the json_schema option and specify response_format:

const response = await browserAI.generateText('List 3 popular colors', {
  json_schema: {
    type: "object",
    properties: {
      colors: {
        type: "array",
        items: {
          type: "object",
          properties: {
            name: { type: "string" },
            hex: { type: "string" }
          }
        }
      }
    }
  },
  response_format: {
    type: "json_object"
  }
});
 
// Returns:
// {
//   "colors": [
//     { "name": "blue", "hex": "#0000FF" },
//     { "name": "red", "hex": "#FF0000" },
//     { "name": "green", "hex": "#00FF00" }
//   ]
// }

Schema Examples

Person Profile Schema

const response = await browserAI.generateText('Create a profile for John', {
  json_schema: {
    type: "object",
    properties: {
      name: { type: "string" },
      age: { type: "number" },
      hobbies: {
        type: "array",
        items: { type: "string" }
      }
    }
  },
  response_format: {
    type: "json_object"
  }
});

Try it Out

You can find a complete working example in the examples/schema_llm directory of the BrowserAI repository. This example demonstrates:

Different schema types and structures
Real-time schema validation
Error handling
UI integration

To run the example:

cd examples/schema_llm
npm install
npm run dev

Best Practices

Always include response_format: { type: "json_object" } when using JSON schemas
Make your prompts specific about the structure you expect
Use clear property names in your schema
Consider adding validation constraints in your schema (e.g., minimum/maximum values)
Handle the response as JSON using JSON.parse()

Text Generation Options Web Worker Support