Information Extraction

Extract structured fields from files or parsed documents using a JSON Schema, retaining source-tracking information where available.

Use it for defined fields such as contract details, receipt values, or resume information. For free-form summaries, use AI Inference.

Define input and fields

Supply either upstream documents or files for this node to parse. Do not supply both.

Setting

Description

Extraction schema

Required JSON Schema defining field names, types, and constraints

Extraction instructions

Optional explanation of field meanings and requirements

Layout-sensitive fields

Fields whose interpretation depends on page position, such as supplier and recipient

Page filter

Limits source pages; leave empty to process all pages

Debug records

Off by default; records some call details for troubleshooting

File input must use valid, unique Catalog file_id values.

Choose an extraction mode

Mode

Results and failure behavior

One combined result, n_to_1

Default; all inputs contribute to one result

One result per source, n_to_n

Produces separate results by source file; supports a failure policy

The default n_to_n policy is per_item: failed sources retain error objects while other sources continue. With fail_fast, a failure stops the batch. Before writing a batch to a table, separate successful results from errors and write only successful data.

Use the output

result is a JSON string for a single result; results contains result strings returned by source. mode identifies the mode, and sources_tracking contains available field-source information.

Distinguish JSON strings from objects when binding to downstream inputs. A single result can feed the JSON input of Save to MOI Table. Process individual results and error states before using multiple results.

Example: extract a contract number

Flow: General Document Parsing → Information Extraction → Save to MOI Table.

Example schema:

{
  "type": "object",
  "properties": {
    "contract_number": {"type": "string", "description": "Contract number"}
  },
  "required": ["contract_number"]
}

Choose combined-result mode and explain what identifies the contract number. After running, verify the extracted number and its source.

Last updated on