Resources · n8n integration

n8n Summarization Chain nodeConfigure Summarization Chain in n8n.

The Summarization Chain node turns a wall of text into a short summary, right inside a workflow. The n8n Summarization Chain node exposes 5 parameters and 3 connection ports, and it leans on a chat model you plug into it. Built for anyone who reads more documents than they have time for.

Try the Summarization Chain node on n8nSee the 5 parameters

Verified Trustpilot reviews · AI, automation & growth agency

Why automate

What does the n8n Summarization Chain node actually do?

It takes text that is too long to read and gives back a condensed version of it. The node description is blunt about it: it transforms text into a concise summary. You feed it the JSON coming from the previous node, a binary file, or documents handed over by a loader sub-node, meaning a small node you attach to a port rather than place on the canvas. The chat model you connect does the actual writing, and the node handles the plumbing around it: splitting long input into chunks, summarizing each one, then combining the pieces.

The first scenario is a reading queue. A workflow scrapes pages, pulls the text out of the HTML, and pushes it into the chain with Data to Summarize set to nodeInputJson. What comes out the other side is short enough to post somewhere a human will see it, in Slack or as a row in Google Sheets. The public n8n template 1951 wires exactly that shape, with an OpenAI Chat Model and a Recursive Character Text Splitter behind it.

The second is document triage. Resumes, contracts and supplier reports arrive as files, a Default Data Loader reads them, and the chain returns a paragraph per document that a person can scan in seconds. Template 2860 does this for candidate evaluation with Google Drive on the front end. Because the summaries are plain text, they drop straight into a sheet or a table without any extra parsing step.

The third is mailbox load. Long threads in Gmail get compressed before anything else happens to them, which keeps a downstream classifier or an AI Agent from burning its context window, the amount of text a model can hold at once, on quoted replies.

Pick a different node when the job is different. If the answer you want is a decision, a label or a structured object rather than a shorter text, a chain built for that, or an agent that can call tools, fits better. If you only need a summary of one short paragraph, a single prompt to OpenAI is lighter than a chain. The Summarization Chain earns its place when the input is longer than a model can swallow in one go, because chunking and combining are exactly what it automates.

The limits are worth knowing before you build. The node summarizes, and nothing else: it does not extract fields, it does not answer questions about the document, and it does not store anything between executions. Quality and cost belong to the model provider you connect, not to n8n. And the version documented here is version 2 of the node, so an older workflow may show fewer options in the panel. For a wider look at the platform, the n8n review covers what it does well and where it hurts.

Connections

What do you plug into the Summarization Chain node?

This node is a root node: it sits on the canvas and receives workflow items on its main input, while sub-nodes attach underneath it, one type per port, and never run on their own.

Ports (what plugs in)

  • Chat Modelai_languageModelrequired
  • Documentai_documentrequired
  • Text Splitterai_textSplitteroptional
01Chat ModelThis is the port that does the writing. Every chunk of text and the final combination pass through the model connected here, so an empty port stops the node before it starts.

This is the port that does the writing. Every chunk of text and the final combination pass through the model connected here, so an empty port stops the node before it starts.

Key parameters

  • Required: the chain cannot produce a summary without a chat model, meaning a sub-node that wraps a text generation model and carries the provider credential.
  • OpenAI Chat Model and Anthropic Chat Model are the usual starting points, and both put the API key on the sub-node rather than on the chain.
  • Ollama Chat Model keeps the text on your own machine, which matters when the documents are contracts or resumes.
Use cases
a nightly digest of scraped articles runs happily on one connected model, and swapping providers later means changing this sub-node only.
02DocumentThe Document port carries the content to summarize when it does not come straight from the previous node's JSON. It is what documentLoader mode reads from.

The Document port carries the content to summarize when it does not come straight from the previous node's JSON. It is what documentLoader mode reads from.

Key parameters

  • Required: the port is part of the node's shape, and the chain expects something on it rather than an empty socket.
  • Default Data Loader is the sub-node the catalog offers here, and it is where binary files become text the model can read.
Use cases
a PDF pulled from Google Drive goes through the loader, then the chain, and lands as a readable paragraph in a spreadsheet.
03Text SplitterA text splitter decides where long content gets cut before the model sees it. This port only matters once you switch Chunking Strategy to advanced.

A text splitter decides where long content gets cut before the model sees it. This port only matters once you switch Chunking Strategy to advanced.

Key parameters

  • Optional: leave it empty and the chain uses the simple settings defined in its own panel.
  • Recursive Character Text Splitter cuts on structure first, which keeps paragraphs intact more often than a raw character count.
  • Token Splitter counts in tokens, the units a model bills and reads in, rather than in characters.
  • Character Text Splitter stays the blunt option when the input has no structure worth respecting.
Use cases
a long transcript with speaker turns survives much better with a recursive splitter than with a fixed cut.
Parameters

Which parameters matter in the Summarization Chain node?

The Summarization Chain node has 5 parameters. For each one: the node as you configure it in n8n, what the parameter changes, and our field notes.

01

Data to Summarize

operationMode

What you see in n8n

Notes & use cases

This selector decides where the text comes from, and the rest of the panel changes with it. Set it first, then configure the rest.

Key parameters

  • Use Node Input (JSON) (nodeInputJson): summarizes the JSON data coming into the node from the previous one, the mode to use after an HTTP Request or an HTML extraction.
  • Use Node Input (Binary) (nodeInputBinary): summarizes the binary data coming in, for a file attached to the item rather than text in a field.
  • Use Document Loader (documentLoader): hands the job to a loader sub-node with more configuration options.
Use cases
a scraping workflow stays on JSON, while a Drive folder full of reports is better served by the loader.
02

Chunking Strategy

chunkingMode

What you see in n8n

Notes & use cases

Long input never reaches the model in one piece. This parameter picks who decides the cut, the node itself or a dedicated sub-node.

Key parameters

  • Simple (Define Below) (simple): the two size fields appear right under it, and nothing else needs connecting.
  • Advanced (advanced): connects a splitter sub-node with more configuration options, on the Text Splitter port.
Use cases
start simple on a first build, then move to advanced once a document type keeps getting cut in the wrong place.
03

Characters Per Chunk

chunkSize

What you see in n8n

Notes & use cases

Available under the simple strategy, this field controls the maximum size, in number of characters, of the final document chunk. Bigger chunks mean fewer calls to the model and more context in each one.

Key parameters

  • Characters Per Chunk (chunkSize): a number, so it accepts an expression such as {{ $json.chunkSize }} when the value depends on the document type flowing through.
Use cases
short support emails do not need the same chunk as a supplier contract, and one workflow can serve both by driving this field from the item.
04

Chunk Overlap (Characters)

chunkOverlap

What you see in n8n

Notes & use cases

Cutting text cleanly in two loses whatever sentence sat on the seam. This field specifies how many characters overlap between chunks, so each piece starts with the tail of the previous one.

Key parameters

  • Chunk Overlap (Characters) (chunkOverlap): a number, kept well below Characters Per Chunk, since overlap is repeated text the model reads twice.
Use cases
legal text and meeting transcripts, where an idea runs across a boundary, are the places overlap actually shows up in the output.
05

Options

options

What you see in n8n

Notes & use cases

Everything that is not the data source or the chunk size sits behind Add Option. Three fields are available here.

Key parameters

  • Input Data Field Name (binaryDataKey): the name of the field in the chain's input that holds the binary file to process.
  • Summarization Method and Prompts: holds the summarization method, with Map Reduce recommended, plus Refine and Stuff, and the Individual Summary Prompts and Final Prompt to Combine you can customize, each of which must keep the "{text}" placeholder.
  • Batch Processing (batching): batch processing options for rate limiting, worth opening when a run pushes many documents at once.
Use cases
rewriting the final prompt is how you get summaries in the house format instead of generic prose.
Need help

Need help automating Summarization Chain with n8n?

A person reads every message.

FAQ

Summarization Chain and n8n, the common questions

01Is the Summarization Chain node included in n8n?
Yes. The node ships in the n8n AI package, @n8n/n8n-nodes-langchain, which comes with n8n itself. There is nothing to install and nothing extra to pay on the n8n side, on n8n Cloud as well as on a self-hosted instance under the Community Edition and its Sustainable Use license. A workflow behaves the same in both places. What does have a cost is the model provider behind the chat model you connect: OpenAI, Anthropic, Google, Mistral and the rest bill their own API calls under their own terms, and n8n adds nothing on top of that.
02What do you need to run the n8n Summarization Chain node?
Nothing on the node itself. The Summarization Chain has no credential and no Authentication selector, so there is no account to connect and no key to paste into it. What it does need is a chat model sub-node on its Chat Model port, and that sub-node carries the provider credential. Connect an OpenAI Chat Model or an Anthropic Chat Model, put the key there once, and the same credential is reusable by other nodes. A required port left empty blocks the execution, so the chain will tell you quickly if the model is missing.
03What are the limits of the Summarization Chain node?
It summarizes, and that is the whole scope. It does not extract structured fields, it does not answer questions about a document, and it keeps no memory between executions. Input longer than the model can handle has to be chunked, which is what Characters Per Chunk, Chunk Overlap (Characters) and the Text Splitter port exist for, and a badly placed cut shows up as a clumsy summary. Quality depends on the model you connect. This page documents version 2 of the node, so an older workflow may expose fewer options in the panel.
04What is the minimum to connect, and which sub-node should a beginner pick?
The minimum is a chat model. Attach one sub-node to the Chat Model port, leave Data to Summarize on Use Node Input (JSON), and the chain already works on text coming from the previous node. An OpenAI Chat Model is the straightforward first pick, since the key setup is short and the model handles plain summarization well. Add a Default Data Loader on the Document port once you start working with files, and only reach for a splitter sub-node, such as the Recursive Character Text Splitter, when the simple chunking settings stop being good enough.
05n8n or Make for summarizing documents?
It depends on where you want the data to live. Make is a hosted platform with no self-hosting option, billed per operation, which suits a team that wants nothing to run on its own servers. n8n runs on n8n Cloud or on your own infrastructure, so confidential documents can be summarized without leaving your network, especially with a locally hosted model. On cost, per-operation billing is predictable for small volumes and less so when a chain splits one document into many chunks. On logic, both are visual, and the sub-node model here is specific to n8n.
Hack'celeration Lab

Get our weekly integration tips.

No spam. Unsubscribe anytime.

Same platform

Other docs you might need on this stack.

Browse all integrations