Create a Knowledge Base
Name a knowledge base and tune how its documents are split into chunks.
Open Automation → Knowledge Base and click New Knowledge Base. The create dialog asks for a name, an optional description, and the chunking settings that control how every document added to this knowledge base is split before it is embedded.
Fields
| Field | What it controls |
|---|---|
| Name | The display name shown in the list and selected from workflow steps. |
| Description | Optional free text to help your team understand what's inside. |
| Min chunk size (characters) | The smallest a chunk may be before it's merged with a neighbor. The create dialog pre-fills 1. |
| Max chunk size (tokens) | The upper bound on a single chunk. The create dialog pre-fills 1024. |
| Overlap (tokens) | How many tokens each chunk repeats from the previous one so context isn't lost at the boundary. The create dialog pre-fills 200. |
You can upload documents directly in the create dialog, or create the knowledge base empty and add documents later.
Choosing chunk settings
Chunking is a trade-off between precision and context:
- Smaller chunks return more precise matches but may cut an idea in half.
- Larger chunks preserve more surrounding context but dilute relevance scores and cost more tokens when passed to a model.
- Overlap softens the boundaries so a sentence split across two chunks still appears in full in at least one of them.
The pre-filled values (min 1 char, max 1024 tokens, overlap 200 tokens) work well for prose
documents such as PDFs, Markdown, and help-center articles. Raise the max chunk size for
long-form reference material where whole sections should stay together; lower it for dense,
fact-dense content such as FAQs.
Chunk settings apply at ingestion
Chunk settings are applied when a document is processed. Changing them affects documents added afterward — re-upload existing documents if you want them re-chunked with the new values.
Tags
Knowledge bases and their documents can be tagged. Tags let you filter the list and, more
importantly, narrow retrieval at query time — an agent can be restricted to documents
tagged policy while ignoring everything else in the same knowledge base. See
Use it in workflows & agents.
How is this guide?
Last updated on