Skip to main content
The current docs are written for the 2.0 structured-output API.

​
What Changed

The public model is now:
  • StructuredOutput for request construction
  • StructuredOutputRuntime for runtime behavior
  • PendingStructuredOutput for lazy execution
  • StructuredOutputResponse as the primary final response object
  • StructuredOutputStream for streaming reads and final stream access

​
Response Ownership

Older docs and examples often treated the raw Polyglot response as the main response object. That is no longer the intended API.
  • use response() when you want the final Instructor response
  • use get() when you want only the parsed value
  • use inferenceResponse() or finalInferenceResponse() only when you need raw transport-level details

​
Streaming Contract

Streaming is now built around Instructor-owned stream state.
  • Polyglot streams deltas
  • Instructor accumulates those deltas in StructuredOutputStreamState
  • final stream reads return StructuredOutputResponse, not raw partial snapshot objects
If you relied on old partial snapshot behavior, update that code to consume:
  • stream()->responses() for partial and final StructuredOutputResponse items
  • stream()->partials() for parsed partial values
  • stream()->sequence() for completed sequence items

​
Runtime Setup

Runtime configuration belongs on StructuredOutputRuntime, not on a global Instructor object.
  • create() returns a lazy handle
  • stream() returns a dedicated stream object
  • StructuredOutput::fromConfig(...) and StructuredOutput::using(...) remain valid entry points
  • published config files are optional

​
Migration Rule

If you are updating older code, rewrite it around one of these shapes:
  • StructuredOutput->with(...)->get()
  • StructuredOutput->with(...)->response()
  • StructuredOutput->with(...)->stream()