structured-output

Topics in the source

• Structured Output
• Quick Start
• When to Use
• Best Practices
• Schema Design
• Mode Selection
• Validation & Recovery
• Dependencies

Getting LLMs to reliably emit valid, schema-constrained output. Data extraction into typed records API responses that must match a contract Tool/function arguments generation Reducing malformed-output handling Define a strict JSON Schema per output shape Use additionalProperties: false to reject surprises Keep schemas flat and primitive where possible Version schemas alongside your code Prefer the provider's responseformat (jsonobject / jsonschema) Use function/tool calling when the output feeds a tool Fall back to few-shot + validation for models without structured mode Never trust unvalidated text Validate every response against the schema Re-prompt with the validation error on failure Cap retries; fall back to a default record Log malformed responses for prompt tuning 1. Define the exact output shape as a JSON Schema. 2. Pick the strictest mode the provider supports (schema > jsonobject > few-shot). 3. Implement extraction with the schema attached. 4. Validate every response; re-prompt with errors on mismatch. 5. Add retry limits and safe defaults. 6. Log failures to improve the prompt or schema. 7. Add tests with realistic inputs including edge cases. 8. Version schemas and mon

This is an excerpt. Refer to the complete source and its current requirements.