An output parser converts a model's raw text reply into structured, usable data.
DefinitionWhat it means
An output parser is the step in an orchestration pipeline that takes a model's raw text response and converts it into a structured object, JSON, a list, a typed field, that the rest of the application can consume. Parsers range from simple regex and delimiter splitting to schema-validating parsers that retry or repair malformed output automatically.
Why it mattersWhy you should care
The output parser is where an orchestration pipeline is most likely to break in production, since a model can drift from the expected format even with a good prompt; robust parsers with validation and automatic retry-on-failure are what separate a demo from a resilient system. It is the direct downstream partner of structured-output prompting, one shapes the request, the other enforces the result.
At a glanceSee it
Opens the parser box — the fixed strip, locate, parse, and type-coerce steps that turn free prose into a typed object.
The upstream design choice — constrain the model to emit valid structure, or let it write freely then parse and repair after the fact.
Where you see itIn the wild
- Parser classes in orchestration frameworks that validate JSON against a schema.
- Retry-on-parse-failure logic wrapping a model call in production pipelines.
- Chains ending in a typed parser step before data reaches the application layer.