README
¶
Structured (JSON) output
Ask an LLM for strict JSON and decode it into a typed Go struct.
model.ChatRequest.ResponseFormat controls the output format:
"json_object"— the model returns valid JSON (no schema enforcement)."json_schema"— the model returns JSON conforming to the schema inMetadata["json_schema"]. Enforced as a native request parameter by OpenAI, Azure OpenAI, OpenAI-compatible providers (Ollama, Mistral, ...), Gemini, and the Responses API. Anthropic has no equivalent parameter, so it falls back to whatever the system prompt asks for.
Going through sdk/agent's WithOutputSchema/output_schema: instead of a
bare Provider.Chat call (as this example does) additionally validates the
reply against the schema after the fact, on every provider — see
Structured (JSON) Output in the
Agents guide.
This example requests a Recipe as JSON, then decodes the reply into the
Recipe struct (tolerating stray Markdown code fences).
Run
Needs a real LLM provider (one network call). Configure any provider from
examples/internal/providers:
OPENAI_API_KEY=sk-... go run ./examples/structured_output/
With no provider set, it prints the required env vars and exits.
Test
The test covers the pure JSON parsing helpers only — no network, no provider:
go test ./examples/structured_output/
Documentation
¶
Overview ¶
Example: structured_output — asking an LLM for strict JSON and decoding it.
What you'll learn:
- How to request JSON output with model.ChatRequest.ResponseFormat
- The difference between "json_object" (valid JSON) and "json_schema" (JSON conforming to a schema in Metadata["json_schema"])
- How to decode the model's reply into a typed Go struct
This example needs a real LLM provider (it makes one network call). Configure one via the environment (OPENAI_API_KEY, ANTHROPIC_API_KEY, GEMINI_API_KEY, …); see examples/internal/providers for the full list. With no provider set it prints the required env vars and exits cleanly.
Run:
OPENAI_API_KEY=sk-... go run ./examples/structured_output/
Source Files