Skip to main content
Constrains every LLM call to a JSON schema using WithResponseFormat — the model returns parseable structured data, not free-form markdown. Source: examples/agent_with_json_response/

​
What it demonstrates

  • interfaces.ResponseFormat with ResponseFormatJSON and JSONSchema
  • Parsing result.Content as JSON after Run
  • Schema enforcement across tool rounds

​
Run

From examples/: 
go run ./agent_with_json_response "What is the capital of France?"

​
Key code

Define the schema once at agent construction: 
rf := &interfaces.ResponseFormat{
    Type: interfaces.ResponseFormatJSON,
    Name: "FactAnswer",
    Schema: interfaces.JSONSchema{
        "type": "object",
        "properties": interfaces.JSONSchema{
            "answer":       interfaces.JSONSchema{"type": "string"},
            "confidence":   interfaces.JSONSchema{"type": "string", "enum": []any{"low", "medium", "high"}},
        },
        "required":             []any{"answer", "confidence"},
        "additionalProperties": false,
    },
}

a, err := agent.NewAgent(
    agent.WithLLMClient(llmClient),
    agent.WithResponseFormat(rf),
)
See Response format for provider support and limitations.

​
Expected output

{"answer":"Paris","confidence":"high"}
The model is constrained to return valid JSON matching the schema — result.Content is always parseable with json.Unmarshal.

​
Learn more

Response Format

Schema design and parsing

Token Usage

Usage on structured runs