TURNS DOCUMENTS INTO DATA THE SMARTER WAY

n8n-nodes-ragextract

The official n8n community node for Ragextract — semantic search and structured outputs for documents.

Built on the Ragextract v2 API. Works as a regular node and as an
AI agent tool.

Please use this project's issue tracker
for bugs and feature requests, or get in touch for help with your
account.

What is Ragextract?

Ragextract helps you get through hundreds of documents quickly and extract facts and figures in
minutes using AI. Under the hood, we deploy a vision-based document question & answer and RAG
pipeline.

You can use Ragextract to…

• Turn a folder of contracts into a table — one row per agreement, one column per question
  (parties, term, renewal notice, liability caps), with every answer citing the page it came from.
• Read supplier invoices that no two suppliers lay out the same way — supplier, invoice number,
  dates, totals, tax and line items, ready to post into your accounting system.
• Watch lease and contract dates — current term ends, notice periods and auto-renewal — so a
  scheduled workflow raises the alert before the window closes.
• Compare subcontractor bids side by side — bid totals, scope exclusions, hourly rates and
  material lead times across every bid for one package.
• Pull diagrams and figures out of technical documents, not just text: image columns answer with
  a region of a page, so charts, floor plans and care pathways come back too.
• Give an AI agent page-level retrieval over thousands of pages, so it answers from the three
  pages that matter instead of stuffing a whole document into the context window.

Learn more at ragextract.com.

Installation

Cloud and self-hosted

Install via Settings → Community Nodes in your n8n instance, searching for
n8n-nodes-ragextract.

Self-hosted CLI

npm install n8n-nodes-ragextract

Credentials

You need a Ragextract account and an API key.

1. Sign up at ragextract.com
2. Go to Settings → Keys → Create API Key
3. Paste the key into the node's Ragextract API credential

Two kinds of key, and what each one gets you

Key	Reaches
psk_… personal key	Every workspace you can, at your own permission level. Recommended.
sk_… workspace key	The single workspace it belongs to.

Both work with this node. A personal key can be narrowed to specific workspaces and tables, with
read or read-write on each; an unscoped one does everything you can do and follows your permissions
if they change. See Permissions and
Authentication.

Base URL only needs changing for a private deployment. Leave it at https://api.ragextract.com
otherwise — the node appends /v2 itself.

Resources and operations

Resource	Operations
Workspace	Get Many, Get, Create, Update
File	Upload, Get Many, Get, Get Items, Get Item, Search, Share, Delete
Job	Get Many, Get, Cancel
Table	Get Many, Get, Create, Update, Delete
Table Column	Get Many, Create, Update, Delete, Run
Table Row	Get Many, Create, Delete
Table Cell	Get Many, Set Override, Clear Override, Get Events, Get Facts
Table Run	Start, Preview, Get Many, Get, Cancel, Cancel All
Bundle	Get Many, Get, Create, Update, Delete, Add Files, Update File, Remove File

Every operation except Workspace → Get Many and Workspace → Create takes a Workspace: v2
nests every resource under one. Pick it from the dropdown, or supply an ID by expression.

How the node behaves

Upload is the only ingest verb. Every upload is parsed and embedded, so a file is searchable
as soon as it finishes. There is no separate vectorize step to remember.

Upload waits by default and returns the finished file, so the next node can use it directly. Turn
Wait For Completion off to get the job back immediately and poll it with Job → Get later —
the right choice for anything large enough that holding the execution open is unreasonable.

Large files just work. Anything over 100 MB switches to a multipart upload automatically, and
rate limits are retried with backoff rather than surfaced as a failure.

Pages come back as binary. Search, Get Item and Get Items have a Download Files
toggle that fetches each result's page through its signed share link and attaches it as binary data —
ready to hand to a vision model or write to storage.

Table runs spend credits, per cell. Table Run → Preview prices exactly the same selection
beforehand and charges nothing. Worth wiring in before a run over a large table. See
Credits.

A rerun never overwrites a human correction, and it skips rows whose documents are still
ingesting — so the cell count it reports is what actually started, and skippedRowIds names the rows
that fell out. See Running and reruns.

Given both rows and columns, a run's selection is their union — every cell of those rows, plus
every cell of those columns.

Recipes

Ingest a document and get its pages

Trigger → Ragextract (File → Upload) → Ragextract (File → Get Items)

Point Upload at a binary property (data) or a public URL. It waits for ingest and outputs the
file, so Get Items can take {{ $json.id }} as its File. Set Page Format to jpg and
Download Files on to get page images out the other side.

Semantic search over a workspace

Trigger → Ragextract (File → Search) → AI Agent

Search returns the matching pages with a relevance score and a signed link each. Leave File
IDs empty to search the whole workspace, or narrow it to specific documents. With Download
Files on, each match arrives with its page attached.

Pull a structured table out of a batch of documents

Ragextract (Table → Create)
  → Ragextract (Table Column → Create)   ← one per question
  → Ragextract (Table Row → Create)      ← one per file
  → Ragextract (Table Run → Preview)     ← what will this cost?
  → Ragextract (Table Run → Start)       ← waits, returns the cells

Each column is a prompt plus an output type (date, number, categorical, boolean,
text_quote, list_scalar, image). Cells come back with the value, a confidence score and
citations pointing at the page and quote they came from.

Set a table to Standing and it runs itself against every new file uploaded to the workspace.

Treat a contract and its amendments as one subject

Ragextract (Bundle → Create) → Ragextract (Bundle → Add Files) → Ragextract (Table Row → Create)

Create the row with Subject Type bundle. Mark the master document Primary and give each
amendment an Effective At date; the extraction resolves them in precedence order and
Table Cell → Get Facts shows what each document contributed. See
Bundles.

Using it as an AI agent tool

The node is available to AI Agent nodes as a tool. File → Search is the obvious one to expose —
it lets an agent look things up across your documents without you building a retrieval chain — but
any operation works, and the agent can fill parameters itself.

Upgrading from 0.1.x

0.2.0 targets the v2 API and is a breaking change. Existing Ragextract nodes need reconfiguring:

0.1.x	0.2.0
(nothing)	Every operation now takes a Workspace
Workspace → Extract / Vectorize	File → Upload (always embeds)
Workspace → Search	File → Search
Dataset → *	File → *
Dataset → Vectorize	gone — upload embeds
{{ $json.data.id }}	{{ $json.id }} — output is no longer wrapped

New in 0.2.0: the Table, Table Column, Table Row, Table Cell, Table Run and Bundle resources,
workspace CRUD, multipart upload, and a configurable base URL.

The v1 API is unchanged and still served; only this node moved.

Development

npm install
npm run dev      # run against a local n8n instance
npm run lint     # n8n community-node lint rules
npm run build    # compile to dist/ and copy icons
npm run verify   # check the build produced what package.json points n8n at

Releasing

Published with npm directly — there is no release tool to drive.

npm version <patch|minor|major>   # bumps package.json, commits, tags
git push && git push --tags
npm publish                       # lint, build and verify run first

prepublishOnly runs lint && build && verify, so a publish always ships a freshly compiled
dist/ and fails rather than uploading one that is missing the node, the credential or their icons.
Update CHANGELOG.md by hand before bumping.

Support

• Issue tracker — bugs and feature
  requests for this node
• ragextract.com/support — account, billing and product help
• support@subworkflow.ai

Licence

MIT

Resources

• Quickstart
• API reference
• Tables · Bundles · Credits · Limits
• Ragextract JS SDK for programmatic usage