Auto strategy

The Auto strategy dynamically routes each page to the Fast, High Res, or VLM partitioning strategy at runtime, optimizing for the highest quality at the lowest cost.

Type: partition

Subtype: vlm

Settings

provider · string

VLM provider to use if Auto routing selects the VLM strategy. Allowed values: anthropic, auto, bedrock, openai, vertexai. Default: anthropic.

provider_api_key · string

If specified, overrides the default API key used for VLM provider calls. Default: none (uses Unstructured’s internal default).

model · string

VLM model to use if Auto routing selects the VLM strategy. Default: claude-sonnet-4-20250514.

For a full list of the models available in Unstructured, see Available models.

output_format · string

Format of the partitioner response. Allowed values: text/html, application/json. Default: text/html.

format_html · boolean

If true, applies Beautiful Soup’s prettify method to any HTML generated by the VLM partitioner, adding indentation for readability. Default: true.

unique_element_ids · boolean

If true, assigns UUIDs to element IDs, guaranteeing uniqueness (useful when used as database primary keys). If false, assigns a SHA-256 hash of the element’s text as its ID. Default: true.

is_dynamic · boolean · required

Must be true to enable dynamic routing of pages to Fast, High Res, or VLM as needed for optimal performance and cost.

allow_fast · boolean · required

Must be true to allow routing of pages to Fast partitioning for better performance and cost savings.

Python SDK

auto_partitioner_workflow_node = WorkflowNode(
    name="Partitioner",
    subtype="vlm",
    type="partition",
    settings={
        "provider": "<provider>",
        "provider_api_key": None,
        "model": "<model>",
        "output_format": "<output-format>",
        "format_html": <True|False>,
        "unique_element_ids": <True|False>,
        "is_dynamic": True,
        "allow_fast": True
    }
)