Resources · n8n integration

n8n Default Data Loader nodeConfigure Default Data Loader in n8n.

Everything an AI node reads has to become a document first, and that is this node's whole job. The n8n Default Data Loader node takes JSON or a binary file from the previous step, cuts it through a text splitter, and hands the pieces to a summarization chain. Six parameters, one required port.

Try the Default Data Loader node on n8n See the 6 parameters

Verified Trustpilot reviews · AI, automation & growth agency

Why automate

What does the n8n Default Data Loader node do in a workflow?

The Default Data Loader is a sub-node, meaning it never runs on its own: you drag it onto a port of a root node and it feeds that node. It reads data from the previous step in the workflow, either the JSON of the incoming item or a binary file attached to it, and converts it into documents, the chunked text format AI nodes expect. Its description in n8n says exactly that: load data from previous step in the workflow.

First scenario, a support knowledge base. A PDF lands in the workflow, Type of Data is set to Binary, Data Format to PDF, and Split Pages in PDF turns each page into its own document so an answer can point at one page instead of a whole manual. The loader passes the result on through its Document output.

Second scenario, records instead of files. Rows pulled from Google Sheets or messages fetched from Gmail arrive as plain JSON. Type of Data stays on JSON, Mode on Load Specific Data, and JSON Pointers keeps only the fields worth indexing, so the body of a ticket gets loaded and the internal identifiers do not.

Third scenario, a spreadsheet export. With Data Format on CSV, CSV Separator matches whatever the export actually used and CSV Column picks the one column that carries the text. Everything else in the file stays out of the index. Add a couple of Metadata properties and retrieval can later filter on them.

When to reach for something else. This node prepares documents, it does not query them and it does not talk to a model. A chat that answers questions needs a root node like AI Agent with a chat model such as OpenAI branched onto it, and the loader only appears on the indexing side of that setup. If your text is already sitting in a field and needs no file parsing at all, the loader in JSON mode with Load All Input Data is the short path.

Known limits. The node exposes six parameters and no more: no chunk inspection, no cleanup step, no deduplication. A text splitter is required, so the chunking decision belongs to the sub-node you connect, not here. And if you set a specific Data Format while the incoming file is something else, the node errors instead of guessing. Teams who want to work through a full pipeline rather than read about it can start with the n8n training, and the n8n review covers how the platform compares on hosting and cost.

Connections

What do you plug into the Default Data Loader ?

n8n splits AI work between root nodes, which sit in the main flow, and sub-nodes, which hang off their ports; the loader takes one sub-node below it and attaches itself to one root node above.

Ports (what plugs in)

Text Splitterai_textSplitterrequired

Output (what it plugs into)

Documentai_document

01Text SplitterThe splitter decides how a long text is cut into chunks before it becomes documents. Without one connected, the loader has nothing to hand the chunking to and the execution stops.

The splitter decides how a long text is cut into chunks before it becomes documents. Without one connected, the loader has nothing to hand the chunking to and the execution stops.

Key parameters

Required: the port has to carry a sub-node for the loader to run.
Recursive Character Text Splitter: the usual starting point, it cuts on paragraph and sentence boundaries before falling back to raw characters.
Character Text Splitter: cuts on one separator you choose, predictable on structured text.
Token Splitter: cuts by tokens, the units a model counts its input in.

Use cases

a long PDF manual goes through the recursive splitter so paragraphs stay whole, while a log export with one record per line suits the character splitter.

02DocumentThis is the output side. The loader does not sit in the main flow; it connects upward, and the documents it produces are consumed by the root node above it.

This is the output side. The loader does not sit in the main flow; it connects upward, and the documents it produces are consumed by the root node above it.

Key parameters

Required: the loader only does something once a root node accepts its documents.
Summarization Chain: the root node that takes this output, reading the documents and producing a summary of them.

Use cases

a weekly report arrives as a file, the loader turns it into documents, and the summarization chain above returns the short version that gets posted to Slack.

Parameters

Which parameter does what?

The Default Data Loader node has 6 parameters. For each one: the node as you configure it in n8n, what the parameter changes, and our field notes.

Type of Data

dataType

What you see in n8n

Notes & use cases

This first choice changes every field below it.

Key parameters

Type of Data: json processes JSON from the previous step, binary a file attached to it.
Mode: allInputData uses everything flowing into the parent agent or chain, expressionData loads a subset, specificField one named binary field.
Data Format: auto detects by mime type, or pick pdfLoader, csvLoader, docxLoader, epubLoader, jsonLoader or textLoader.
Data: the JSON content, dragged from the input pane or written as {{ $json.body }}.
Input Data Field Name: the field holding the binary file.

Use cases

invoices attached to incoming mail go through Binary with Data Format on PDF, while ticket text already present in the item goes through JSON.

JSON Pointers

options.pointers

What you see in n8n

Notes & use cases

Loading a whole JSON object usually drags in noise. This option narrows the load to named paths inside the object, so only the parts you point at become document text.

Key parameters

JSON Pointers: pointers to extract from JSON, written as /text, or several at once as /text, /meta/title.

Use cases

an article payload carries a title, a body and a pile of tracking fields; pointing at the title and the body keeps the documents readable and leaves the rest behind.

CSV Separator

options.separator

What you see in n8n

Notes & use cases

CSV files are only nominally standard. Exports built in a European locale often use a semicolon where the loader expects a comma, and every row then collapses into a single unreadable cell.

Key parameters

CSV Separator: the separator to use for CSV, set to whatever character the file actually uses.

Use cases

a product export coming out of an accounting tool uses semicolons, so the separator is set to match before anything downstream tries to read a column.

CSV Column

options.column

What you see in n8n

Notes & use cases

Most spreadsheets hold one column of prose and a dozen columns of identifiers, dates and totals. Naming that one column keeps the numbers out of the documents.

Key parameters

CSV Column: the column to extract from the CSV, given by its header name.

Use cases

a survey export has a free text answer next to eight rating columns; extracting the answer column alone gives documents that read like sentences instead of a grid.

Split Pages in PDF

options.splitPages

What you see in n8n

Notes & use cases

A PDF can arrive as one long block of text or as one document per page. This toggle decides which, and it changes how precise anything reading the result can be.

Key parameters

Split Pages in PDF: whether to split PDF pages into separate documents.

Use cases

for a contract where a clause needs to be traced back to where it sits, per page documents keep that link; for a two page memo read end to end, the split adds nothing.

Metadata

options.metadata

What you see in n8n

Notes & use cases

Documents on their own carry no context about where they came from. Metadata attaches properties to each one, and those properties are what filtering during retrieval works on later.

Key parameters

Metadata: metadata to add to each document, added property by property with the Add property button.
Name: the property name, required for each entry.
Value: the property value, often an expression such as {{ $json.source }}.

Use cases

tagging each document with its language and its department lets a later retrieval step answer from the right subset instead of the whole index.

Need help

Need help automating Default Data Loader with n8n?

A person reads every message.

FAQ

Default Data Loader and n8n, common questions

01Is the n8n Default Data Loader node included for free?

Yes. It 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 running the Community Edition under the Sustainable Use licence. A workflow behaves the same in both places, so a setup built on Cloud moves to your own server without rewriting the loader. The only cost that can appear in this kind of workflow comes from elsewhere: whichever model provider you connect further along bills its own API calls under its own terms, and n8n adds nothing to that.

02What credentials does the Default Data Loader need?

None. The node has no credential and no Authentication selector, so there is no account to connect and no key to paste before it runs. That is unusual among AI nodes and it is worth knowing: the authentication lives on the model sub-node connected further along the chain, which carries the provider key. What the loader does need is a Text Splitter on its required port, and data actually arriving from the previous step. If the execution fails here, the cause is a missing splitter, an empty required field or a file whose type does not match Data Format, never a login problem.

03What are the limits of this node?

It prepares documents and stops there. It does not query anything, it does not embed, and it holds no model. Version 1 of the node exposes six parameters, so the room to intervene is deliberately narrow: no preview of the chunks it produced, no cleanup pass, no deduplication between runs. Chunking is delegated to the splitter you connect, which means the chunk settings live on that sub-node and not in this panel. One behaviour surprises people: set a specific Data Format and send a file whose mime type does not match, and the node errors rather than falling back.

04What has to be connected at minimum, and which sub-node comes first?

One sub-node below and one root node above. Below, the Text Splitter port is required, and the Recursive Character Text Splitter is the sensible first choice because it tries to cut on natural boundaries before resorting to raw character counts. Above, the Summarization Chain accepts the Document output. Once that triangle runs, changing one piece at a time is easy: swap in the Token Splitter if a model keeps rejecting oversized input, or the Character Text Splitter when the text has an obvious separator of its own.

05n8n or Make for loading documents into an AI workflow?

It depends on where the documents can live. n8n self hosts through Docker or npm, or runs on n8n Cloud, so a contract or an export can be parsed on a machine you control. Make is hosted only, with no self hosting option, and it bills per operation, which matters when a single file turns into many documents. Visual logic differs too: n8n wires sub-nodes onto ports of a root node, a shape you either find natural or do not. Try the loader on one real file in both and the choice usually makes itself.

Hack'celeration Lab

Get our weekly integration tips.

No spam. Unsubscribe anytime.