Dmitry Ng 4b2f76720e fix(llms): correct reasoning, structured output, and callback edge cases ...

- Honor the user's temperature when reasoning is explicitly disabled and remove bare qwen3 from the reasoning-model catalog.
- Compile structured-output schemas in preflight (ErrStructuredOutputConfig) and keep the raw schema off the public openai.ResponseFormat type without changing the wire.
- Gate Bedrock adaptive thinking by Claude generation, add Sonnet 5 / Fable 5 to prompt caching, and fix Gemini reasoning-off (OffOmit) and Vertex streaming finish reason.
- Unify adapter callbacks to emit exactly one closing event on structured-output/refusal failures.

2026-07-23 12:09:13 +03:00

..

doc.go

feat(llms): add provider-neutral structured output via WithStructuredOutput

2026-07-17 04:39:05 +03:00

schema_test.go

feat(llms): add provider-neutral structured output via WithStructuredOutput

2026-07-17 04:39:05 +03:00

schema.go

feat(llms): add provider-neutral structured output via WithStructuredOutput

2026-07-17 04:39:05 +03:00

validator_test.go

feat(llms): add provider-neutral structured output via WithStructuredOutput

2026-07-17 04:39:05 +03:00

validator.go

fix(llms): correct reasoning, structured output, and callback edge cases

2026-07-23 12:09:13 +03:00