Enhancement nodes - Unstructured

Chunker

These chunker nodes apply AI-powered enhancements to existing chunks rather than splitting content by character or structure.

Node	Description
Contextual chunking	Prepends LLM-generated context summaries to each chunk, improving retrieval quality for ambiguous content.
Chunk by Table Merging	Detects and merges tables that span adjacent pages into a single coherent element using LLM assistance.

Workflow placement

Place the Contextual chunking node after one of the standard text chunkers (such as chunk_by_character, chunk_by_page, chunk_by_similarity, or chunk_by_title), as it enriches existing chunks rather than creating them.
The Chunk by Table Merging node pairs well with a Table to HTML enrichment earlier in the workflow.

Embedder

Configure an Embedder node to convert chunked document elements into vector embeddings for similarity search in vector stores.

Extract

Learn about the available extraction methods.

Node	Description
LLM	Extracts structured data using a language model with a JSON schema or plain-language guidance.
Regex	Extracts structured data using named regular-expression patterns, without requiring a language model.

Prompter

Unstructured can potentially generate image summary descriptions, table summary descriptions, table-to-HTML output, and generative OCR optimizations, only for workflows that are configured as follows:

With a Partitioner node set to use the Auto or High Res partitioning strategy, and an image summary description node, table summary description node, table-to-HTML output node, or generative OCR optimization node is added.
With a Partitioner node set to use the VLM partitioning strategy. No image summary description node, table summary description node, table-to-HTML output node, or generative OCR optimization node is needed (or allowed).

Even with these configurations, Unstructured actually generates image summary descriptions, table summary descriptions, and table-to-HTML output only for files that contain images or tables and are also eligible for processing with the following partitioning strategies:

High Res, when the workflow’s Partitioner node is set to use Auto or High Res.
VLM or High Res, when the workflow’s Partitioner node is set to use VLM.

Unstructured never generates image summary descriptions, table summary descriptions, or table-to-HTML output for workflows that are configured as follows:

With a Partitioner node set to use the Fast partitioning strategy.
With a Partitioner node set to use the Auto, High Res, or VLM partitioning strategy, for all files that Unstructured encounters that do not contain images or tables.

Unstructured never produces generative OCR optimizations for workflows with a Partitioner node set to use the Fast partitioning strategy.

Learn about the available enrichments.

Node	Description
Image Description	Generates text summaries of detected images using a VLM.
Table Description	Generates text summaries of detected tables using a VLM.
Table to HTML	Converts detected tables to HTML markup using agentic AI or a VLM.
Named Entity Recognition (NER)	Identifies entities, their types, and relationships within text.
Generative OCR	Improves the fidelity of initially partitioned text blocks using a VLM.

Workflow placement

Always place enrichment nodes after a Partitioner node.
Place enrichment nodes before Chunker nodes, unless otherwise noted.
- For example, a table2html enrichment is often paired with a chunk_by_table_merging node later in the workflow.
- An exception is the NER enrichment, which should be placed after any Chunker nodes.
Always place any image description enrichment after any OCR enrichment, if present, so that it overwrites the results from the OCR enrichment.

Chunk by Similarity Contextual chunking

​Chunker

​Workflow placement

​Embedder

​Extract

​Prompter

​Workflow placement

Chunker

Workflow placement

Embedder

Extract

Prompter

Workflow placement