Skip to main content

Structured Output (JSON Schema)

Constrain the agent's final response to a JSON schema by setting output_format. The structured result is available on ResultMessage.structured_output.

Basic example​

from agentix import AgentixAgentOptions, AgentixClient, ResultMessage

schema = {
    "type": "json_schema",
    "json_schema": {
        "name": "analysis_result",
        "strict": True,
        "schema": {
            "type": "object",
            "properties": {
                "summary": {"type": "string"},
                "sentiment": {"type": "string", "enum": ["positive", "negative", "neutral"]},
                "confidence": {"type": "number", "minimum": 0, "maximum": 1},
            },
            "required": ["summary", "sentiment", "confidence"],
            "additionalProperties": False,
        },
    },
}

options = AgentixAgentOptions(
    name="analyst",
    provider="openai",
    output_format=schema,
)

async def main():
    async with AgentixClient(options) as client:
        async for msg in client.query("Analyse: 'I love this product!'"):
            if isinstance(msg, ResultMessage):
                result = msg.structured_output
                print(result["sentiment"])    # "positive"
                print(result["confidence"])   # e.g. 0.95

Accessing structured output​

async for msg in client.query("..."):
    if isinstance(msg, ResultMessage):
        if msg.structured_output is not None:
            # Parsed JSON object matching your schema
            data = msg.structured_output
        else:
            # Fallback plain text
            text = msg.result

Effort / thinking​

For reasoning-intensive structured extraction, combine with effort:

options = AgentixAgentOptions(
    output_format=schema,
    effort="high",   # maps to 10k thinking budget tokens (Anthropic)
)

Provider support​

Structured output is supported on all providers. Implementation details:

  • Anthropic — uses tool_use mode internally for constrained generation
  • OpenAI — uses the response_format parameter with json_schema type
  • Gemini — uses response_schema generation config
  • DeepSeek — uses JSON mode