Output Parsers

Structured extraction from LLM text responses.

`StrOutputParser`

Strip whitespace from the LLM response and return it as a plain string.

`JSONOutputParser`

Parse a JSON value from LLM text.

Strips Markdown fenced code blocks (json ... ) before parsing so the LLM can wrap its JSON in a code block without breaking the parser.

`RegexParser`

Extract named capture groups from LLM text using a regex pattern.

Parameters:

Name	Type	Description
`pattern`	`str`	A regular expression string containing at least one named group `(?P<name>...)`.

`CommaSeparatedListParser`

Parse a comma-separated list from LLM text.

`MarkdownCodeBlockParser`

Extract the first fenced code block from LLM text.

Parameters:

Name	Type	Description
`language`	`str`	Optional language tag to match (e.g. `"python"`). When provided the parser first attempts to find a block with that language tag before falling back to any fenced block.

`PydanticOutputParser`

Parse LLM text as a Pydantic model instance.

Strips Markdown fenced code blocks before JSON-decoding, then validates the resulting dict against the provided Pydantic model class.

Usage:

python

class UserInfo(BaseModel):
    name: str
    age: int

parser = PydanticOutputParser(model=UserInfo)
user = parser.parse('{"name": "Alice", "age": 30}')

Parameters:

Name	Type	Description
`model`	`type[T]`	The Pydantic model class to validate against.

`RetryOutputParser`

Wrap any OutputParser and automatically retry with error feedback when parsing fails.

On each failure a correction turn is appended to the conversation and the LLM is asked to produce a valid response. Retries continue up to max_retries additional attempts before MaxRetryError is raised.

Usage:

python

parser = RetryOutputParser(
    parser=PydanticOutputParser(model=UserInfo),
    llm=llm_service,
    max_retries=3,
)
result = await parser.parse_with_retry(
    original_messages=messages,
    completion=completion,
)

Parameters:

Name	Type	Description
`parser`	`Any`	The underlying parser to wrap.
`llm`	`Any`	The `LLMService` to call for correction turns.
`max_retries`	`int`	Maximum number of additional retry attempts after the first failure. Defaults to `3`.

`MaxRetryError`

Raised when RetryOutputParser exhausts its retry budget.

Parameters:

Name	Type	Description
`message`	`str`	Human-readable description of the exhaustion.
`attempts`	`int`	Total number of parse attempts made.