Reducto Block

The Reducto block connects to the Reducto document AI API to parse documents into structured content, extract data against a JSON schema, and split documents into named sections. Reach for it when you need to process PDFs or other documents into machine-readable output without building your own document-parsing pipeline.

Overview

When to Use

• Convert a PDF or document URL into structured text chunks for downstream LLM or search workflows
• Pull specific fields (e.g. totals, names, dates) from a document using a JSON schema extraction
• Split a multi-section document (e.g. a contract or report) into labelled segments based on natural-language category descriptions
• Automate invoice, receipt, or form processing where the output must conform to a known data shape
• Pre-process documents before storing results in a knowledge base or table block

Configuration

Operation

Dropdown — selects which Reducto API endpoint to call. Required.

Choosing an operation controls which additional fields appear below.

Document URL

Short text input — the publicly accessible URL of the document to process (PDF or other Reducto-supported format). Required for all three operations.

Example: https://example.com/document.pdf

Extraction Schema

(Visible only when Operation is reducto_extract)

Long text input — a JSON Schema object describing the fields Reducto should extract from the document.

Example: {"type":"object","properties":{"total":{"type":"number"}}}

System Prompt

(Visible only when Operation is reducto_extract)

Long text input — an optional natural-language instruction guiding the extraction model (e.g. Be precise and thorough.). Optional.

Split Description

(Visible only when Operation is reducto_split)

Long text input — a JSON array of category objects, each with a name and description, telling Reducto how to segment the document.

Example: [{"name":"invoice","description":"Pages containing an invoice"}]

API Key

Short text input (password-masked) — your Reducto API key. Required for all operations.

Use the reference {{REDUCTO_API_KEY}} to pull this from an environment variable rather than hard-coding the secret.

Inputs & Outputs

Inputs:

• operation (string) — the selected operation id (reducto_parse, reducto_extract, or reducto_split)
• apiKey (string) — Reducto API key
• documentUrl (string) — URL of the document to process
• schema (json) — JSON schema describing fields to extract (Extract operation only)
• systemPrompt (string) — optional extraction guidance prompt (Extract operation only)
• splitDescription (json) — array of category definitions for splitting (Split operation only)

Outputs:

• data (json) — the primary result object returned by Reducto (parsed chunks/blocks, extracted field values, or split sections depending on the operation)
• metadata (json) — job metadata including jobId (Reducto job identifier) and status (HTTP status code)

Tools

• Reducto Parse (reducto_parse) — calls POST https://platform.reducto.ai/parse with the document URL; returns parsed content as structured chunks and blocks with usage statistics.
• Reducto Extract (reducto_extract) — calls POST https://platform.reducto.ai/extract with the document URL, a JSON schema, and an optional system prompt; returns extracted structured data and citations aligned to the schema.
• Reducto Split (reducto_split) — calls POST https://platform.reducto.ai/split with the document URL and a split-description array; returns the document divided into named sections with detected page ranges.

YAML Example

reducto_1:
  type: reducto
  name: "Reducto"
  inputs:
    operation: "reducto_extract"
    documentUrl: "{{trigger.documentUrl}}"
    schema: '{"type":"object","properties":{"invoice_total":{"type":"number"},"vendor":{"type":"string"}}}'
    systemPrompt: "Be precise and extract only values explicitly stated in the document."
    apiKey: "{{REDUCTO_API_KEY}}"
  connections:
    outgoing:
      - target: next-block-id