KB Index Node
Your product list lives in another system and changes every day. Your agent answers questions from a Knowledge Base. The KB Index node connects the two: it takes data from an earlier step, or fetches it from a URL, and indexes it into a Knowledge Base folder so agents can search it.
You'll find it under Data in the node picker.
It's the storing counterpart of Document Processing: Document Processing reads a document once and keeps nothing, while KB Index adds content to a Knowledge Base folder where it stays searchable.
Options

| Field | What it does | Options | Default | Shown when |
|---|---|---|---|---|
| Source | Where the data to index comes from. Use "variable" to index an earlier HTTP Request node's body. | From a previous step (variable) · Fetch a URL | variable | |
| Source value | A {{variable}} holding the data (e.g. an HTTP Request body), or an https URL to fetch. Internal/private addresses are blocked. | |||
| Fetch headers | Request headers when Source = URL (e.g. an auth key). Values support {{variables}}. | Source is Fetch a URL | ||
| Target KB folder | The Knowledge Base folder to index into. A KB Search over this folder will find the indexed items. | |||
| Index mode | csv_rows: a 2-column name,id CSV → one searchable vector per row (each row an independent record). document: normal prose chunking. | Document (normal chunking) · CSV catalog — one record per row | document | |
| File name | Display name for the indexed file(s) in the KB. | |||
| Output Variable | Where to store { indexed_count, kb_record_ids }. Defaults to kb_index_result. |
When Source is Fetch a URL, Fetch headers carries anything the feed needs, such as an API key. A credential header is stored encrypted: once saved it reads Secret set — stored encrypted, not shown with Replace, and it's never shown again or included in copies, templates or run details.
Which Index Mode?
- Use CSV catalog, one record per row for a lookup list: products and their codes, tests and their ids, branches and their numbers. Each row becomes its own searchable record, so a search for one product returns that row and its id.
- Use Document (normal chunking) for prose: policies, descriptions, FAQs.
Large CSV files are split into parts of about 900 rows, each shown in the folder as its own file (Test catalog.csv (1), (2), …).
Re-runs replace, not duplicate
Each time the node runs, it first removes the files it indexed into that folder on its previous run, then indexes the new data. A scheduled workflow therefore keeps one fresh copy of the feed instead of piling up duplicates. Files you uploaded to the folder yourself aren't touched.
What you get back
The Output Variable holds indexed_count (how many pieces were indexed), kb_record_ids, and parts (how many files were created). If the source is empty, the download fails or indexing fails, the run continues and the variable carries an error instead. The step is marked as an error, so the run shows Completed with errors and a scheduled run alerts your workspace.
Worked example: a daily test catalog at Acme Diagnostics
Setup. Acme's lab system publishes its test list as a CSV of test names and codes. Asha builds a workflow: Schedule trigger (daily) → HTTP Request (GET the CSV) → KB Index with Source From a previous step, Source Value {{ http_response.body }}, Target KB Folder Test catalog, and Index Mode CSV catalog.
Action. Every morning the workflow runs and re-indexes the list.
Result. When Priya asks the booking agent "Do you do a vitamin D test?", the agent finds the matching row in the Test catalog folder, with its current code, and answers from it.
What just happened. Each row became a separate searchable record, and yesterday's copy was replaced by today's.
Make sure the agent that should use the data has that folder selected on its Knowledge sub-node.
Next steps
- HTTP Request: fetch the data to index.
- Knowledge Base overview: folders, files and how agents search them.
- Knowledge sub-node: give an agent access to the folder.