A knowledge base is what your AI steps answer from. You point it at web pages or a whole site; Natiq reads them on a schedule, distils what matters into a map of topics, and makes the text searchable so answers draw on your content.
Knowledge has its own entry on the rail. Anyone with builder.bots:read can browse; creating and editing needs builder.bots:write.
Create a knowledge base
Open Knowledge and select New knowledge base.
Enter a Name, and optionally a description for your team.
Under Purpose, say what the assistant needs to know, for example Answer customer questions about our billing, refunds and subscription plans. Pages are distilled to that purpose and nothing else. It is required before a crawl can run.
Under Extraction model, pick a Provider, Connection and Model. It reads each changed page twice, so choose a small, cheap model.
Under Retrieval backends, turn on Voyage AI embeddings, OpenAI vector store, or both. Anthropic and DeepSeek steps read Voyage; OpenAI steps read the OpenAI store. Turn on whichever your steps use.
Select Create.
Optional Chunking and limits are covered in Knowledge map and settings.
Add a source
Open the knowledge base and select Add source on the Sources tab. Sources are web pages or sites.
| Field | What it does |
|---|---|
| Start URL | The page the crawl starts from. Use an absolute http or https address. |
| Follow links | How many link hops from the start page. 0 crawls only that page. Default 2, up to 5. |
| Page cap per crawl | Stops fetching after this many pages each crawl. Default 50, up to 500. |
| Allowed hosts | Other hosts the crawler may follow links into. Empty stays on the start page's host. |
| Crawl every | 15 minutes, hour, 6 hours, day (default), week, or Custom (at least 15 minutes). |
The drawer sums it up, for example Every day, up to 50 pages within 2 link hops of the start page.
Set up before you crawl
If the knowledge base has no purpose, no extraction model or no retrieval backend, the Sources tab warns that crawls will fail or store nothing. Fix it in Settings first.
Manage sources
Each source row shows its depth and pages, hosts, frequency, status and last crawl. Status is Active, Paused, Queued, or a count of failures.
- 1Add source
- 2Status
- Run now queues a crawl, which starts on the crawler's next pass. It needs
builder.bots:trigger. - Pause and Resume stop and restart scheduled crawls. Like Run now, they need
builder.bots:trigger. - Pages shows each fetched page, whether it was extracted and which topic it landed in.
- Edit and Delete. Deleting a source keeps the pages it added in the map until they are re-crawled or deleted.
On the knowledge list, Status reads Needs setup, Failing, No sources, Crawling, First crawl pending, Ready or Nothing extracted, with a hint saying what to do.
Use it in an agent
In an AI Task or AI Agent step, open Knowledge and pick the Knowledge base. The list shows how many sources and topics each has and warns if its backend does not match the step's provider. See AI steps.
Sources today
This version crawls web pages and sites. Only content on the web can be added as a source.