structured-output

To get reliable JSON parsing from an LLM agent, you provide an output_schema so the agent returns a validated Pydantic object instead of free-form text. The system appends parser format instructions, extracts JSON, and validates it against your schema.

Schema validation on streaming responses works with astream() to inspect the final typed object once the response completes. You supply a Pydantic output_schema, and the parser extracts and validates the JSON chunks as they arrive.

To use Pydantic to extract structured data from multi-agent workflows, configure your LlmAgent with a defined output_schema. This enforces consistent structured data extraction across multi-agent compositions by validating model responses into typed objects.

Structured output handles malformed JSON by providing a fallback structured parsing path. If standard JSON extraction fails, the fallback path attempts to recover and validate the response against your Pydantic schema.