class AutoSplitter(BaseSplitterNode):
"""Routes each Document to the most suitable splitter based on metadata and content."""
component_cls: ClassVar[type] = AutoSplitterComponent
name: str = "AutoSplitter"
description: str = "Routes documents to structure-aware splitters and falls back to recursive splitting."
rules: list[AutoSplitterRule] = Field(
default_factory=_default_rules,
description="Serializable rules used to select splitter strategies.",
)
fallback_strategy: AutoSplitterStrategy = Field(
default=AutoSplitterStrategy.RECURSIVE_CHARACTER,
description="Strategy used when no rule or inference matches.",
)
fallback_on_error: bool = Field(default=True, description="Fallback to fallback_strategy on inferred route errors.")
infer_from_content: bool = Field(default=True, description="Infer strategy from lightweight content sniffing.")
add_splitter_metadata: bool = Field(default=True, description="Stamp selected splitter strategy into metadata.")
splitter_metadata_key: str = Field(default="splitter_strategy", description="Metadata key for selected strategy.")
refine_structured_chunks: bool = Field(
default=True,
description="Recursively split oversized Markdown/HTML sections while retaining section metadata.",
)
repair_flattened_markdown: bool = Field(
default=True,
description="Restore line starts before inline Markdown headings in otherwise flattened documents.",
)
flattened_markdown_max_heading_level: int = Field(
default=4,
ge=1,
le=6,
description="Deepest inline Markdown heading level restored in flattened documents.",
)
json_max_chunk_size: int = Field(default=2000, gt=0, description="Maximum serialized JSON chunk size.")
json_min_chunk_size: int | None = Field(default=None, description="Minimum serialized JSON chunk size.")
json_convert_lists: bool = Field(default=False, description="Convert JSON lists to indexed dicts before splitting.")
markdown_headers_to_split_on: list[tuple[str, str]] | None = Field(
default=None,
description="Pairs of (markdown-prefix, metadata-key) for MarkdownHeaderSplitter.",
)
markdown_strip_headers: bool = Field(default=True, description="Drop Markdown header lines from chunks.")
markdown_return_each_line: bool = Field(default=False, description="Emit one Markdown chunk per non-empty line.")
html_headers_to_split_on: list[tuple[str, str]] | None = Field(
default=None,
description="Pairs of (html-tag, metadata-key) for HTML splitters.",
)
html_return_each_element: bool = Field(default=False, description="Emit one HTML chunk per element.")
html_xpath_filter: str | None = Field(
default=None, description="Optional XPath used to scope HTML section splitting."
)
code_parser: CodeParser = Field(default=CodeParser.REGEX, description="Parser to use for code splitting.")
code_default_language: Language = Field(
default=Language.PYTHON, description="Fallback language for code splitting."
)
def _component_kwargs(self) -> dict[str, Any]:
kwargs = super()._component_kwargs()
kwargs.update(
rules=self.rules,
fallback_strategy=self.fallback_strategy,
fallback_on_error=self.fallback_on_error,
infer_from_content=self.infer_from_content,
add_splitter_metadata=self.add_splitter_metadata,
splitter_metadata_key=self.splitter_metadata_key,
refine_structured_chunks=self.refine_structured_chunks,
repair_flattened_markdown=self.repair_flattened_markdown,
flattened_markdown_max_heading_level=self.flattened_markdown_max_heading_level,
json_max_chunk_size=self.json_max_chunk_size,
json_min_chunk_size=self.json_min_chunk_size,
json_convert_lists=self.json_convert_lists,
markdown_headers_to_split_on=self.markdown_headers_to_split_on,
markdown_strip_headers=self.markdown_strip_headers,
markdown_return_each_line=self.markdown_return_each_line,
html_headers_to_split_on=self.html_headers_to_split_on,
html_return_each_element=self.html_return_each_element,
html_xpath_filter=self.html_xpath_filter,
code_parser=self.code_parser,
code_default_language=self.code_default_language,
)
return kwargs