Data Extractor
Like AI::Chat, but for turning documents into structured data. The node reads
one or many files from the file input, feeds their textual content to the
configured AI provider (each document prefixed with its filename so the model
can tell them apart), and forces the reply to conform to the configured JSON
Schema. The reply is parsed and validated against that schema before being
emitted as JSON on the response output — a reply that fails to parse or
validate errors the step.
Pre-requisite: Install an AI provider application from Profile > {Organization} > Applications.
Parameters
Configured AI application that supplies the API credentials and provider type.
Model identifier from the selected provider. Must support structured / JSON-schema output. The form suggests common identifiers based on the selected provider — click a badge to fill in. Defaults to the provider's recommended model when left empty.
System prompt injected before the document content. Use it to tell the model what to extract and how to map the documents onto the schema.
JSON Schema the extracted data must conform to. It is sent to the provider
to constrain generation and used to validate the reply. The schema itself
is checked for validity at configuration time, so a malformed schema fails
the node build rather than a run. An empty schema ({}) accepts any
object — the editor shows a warning while the schema is empty. Two starter
templates are available: Empty schema and Invoice extraction (a worked
example mapping an invoice onto structured fields).
Input
The document(s) to extract from. Accepts a single File or a Files
collection. When several files are provided, their contents are
concatenated into one prompt, each prefixed with its filename.
Output
The extracted data as Json, guaranteed to validate against the
configured schema.