Indexer

Save text from a workflow into a knowledge base, so agents can search it later.

What it does

Indexer splits Input into chunks with the connected text splitter, then saves the chunks into the knowledge base connected to Knowledge. Each chunk is embedded with that knowledge base’s own embedding model.

With a Local Knowledge store, indexing the same source again updates it instead of adding a copy: its chunks are overwritten, and if the new text is shorter, the extra old chunks are removed. Empty chunks are skipped.

Indexer is an end point: it has no output port, so nothing runs after it on its branch.

When to use it

• Fill a knowledge base from a workflow: pages you scrape, files you load, results of an API call.
• Keep a knowledge base up to date on a schedule, by re-indexing the same pages.

To add files, pages or text by hand, use Add sources on the knowledge bases page instead. To ask questions about indexed content, use RAG Agent or Q&A Agent.

Inputs and settings

Réglage	Type	Ce que cela fait
Input OBLIG.(obligatoire)	Text	The input to index.
Source ID	Text	A stable id for this document. Indexing again with the same id updates it instead of adding a copy. Defaults to the source link or name, else to the text itself.
Source name	Text	Shown in citations and in the knowledge base source list. Indexing again with the same source name (or link) updates that source in place instead of adding a copy.
Source link	Text	Where the text comes from, linked from citations.

Fill Input with an expression such as {{ $input.text }}.

A source is identified by its Source ID, else its Source link, else its Source name. With none of them, the text itself is the identity, so a changed text counts as a new source. When a workflow re-indexes the same pages, always set one of the three. Citations show Source name, or the link, or “Workflow input” when both are empty.

Outputs

This node has no output port: nothing runs after it on this branch.

Each output item has these fields:

Field	Type	What it holds
success	True/false
chunks	Number
added	Number
updated	Number
failed	Number
removed	Number
sourceId	Text

Shape of an output item

{
  "success": true,
  "chunks": 0,
  "added": 0,
  "updated": 0,
  "failed": 0,
  "removed": 0,
  "sourceId": "…"
}

The result is visible in the run data:

• chunks: how many chunks the input was split into.
• added and updated: new chunks, and chunks that replaced the same source’s previous version.
• failed: chunks that could not be saved. success is false when any failed.
• removed: old chunks deleted because the source got shorter.
• sourceId: the source the chunks are stored under. Index again with the same id to update it.

After a run, the node on the canvas shows a badge such as 24 added · 3 updated. With a vector store other than Local Knowledge, chunks are simply added and updated is always 0.

Dependencies and credentials

Slot	Accepts	Connections
Model	chat model	One node.
Knowledge OBLIG.(obligatoire)	vector store	One node.
Text Splitter OBLIG.(obligatoire)	text splitter	One node.

• Knowledge: a Local Knowledge node pointing at the knowledge base to fill.
• Text Splitter: how the text is cut, for example Recursive Character Text Splitter.
• Model: not used for indexing. Workflows that already have one connected keep working.

Example workflow

1. Schedule runs every day.
2. Get All Text from Link with URL https://example.com/pricing and Output field name text.
3. Indexer with Input {{ $input.text }}, Source name Pricing page and Source link https://example.com/pricing. Connect Local Knowledge to Knowledge and Recursive Character Text Splitter to Text Splitter.

Each day the same source is updated in place, and chat citations link back to the page.

Common issues

• “None of the N chunks could be saved to the knowledge base.” Nothing was written. Check that the knowledge base exists and that its embedding model is installed in Local models.
• “N of M chunks could not be saved to the knowledge base.” (a warning in the run log) Part of the text was saved; success is false and failed counts the rest. Run again to retry.
• Duplicates appear after each run. No stable identity is set. Fill Source ID, Source link or Source name.
• Citations read “Workflow input”. Set Source name.
• “Vector store not found” or “Text splitter not found”. The node on that slot is not set up. Open it and complete its settings.

Related nodes

• Local Knowledge: the knowledge base Indexer writes to.
• RAG Agent and Q&A Agent: answer questions from what you indexed.
• File Document Loader and URL Document Loader: load documents to index.