Knowledge base

The Knowledge tab is where you control what your agent knows. Knowledge lives in knowledge bases — collections of crawled pages, uploaded documents, and text entries — and the agent searches the knowledge bases assigned to it live during a conversation, so it answers from your real content instead of guessing.

How it works

  1. You connect a website. Seny crawls it — following internal links from the root URL. The crawl is capped at roughly 80 pages by default (the crawler ceiling); the dashboard requests up to 500, which the crawler service clamps down.
  2. Each page is cleaned, summarized, and embedded into a searchable index, alongside any documents and text entries in the same knowledge base.
  3. When a visitor asks something, the agent runs a knowledge search and pulls the most relevant content into its answer.

This all happens automatically — there's no "train the model" step. Connect the site, let the crawl finish, and the agent can answer.

Connecting a site

When you create an agent, Seny connects and crawls the site you signed up with. The connect-site card and live crawl progress — the count of pages indexed so far (plus failed and skipped), shown with an indeterminate progress bar rather than a "pages total" figure — live on the agent's Status tab. Sites, documents, and other sources are managed under Knowledge bases in the dashboard.

The agent's Knowledge tab is where you pick which knowledge bases the agent draws from, browse the crawled pages in each, add agent-specific text entries, and test retrieval with a live search.

Keeping it fresh

Each site has a re-crawl cadence — off, daily, weekly, or monthly — settable via the API. Seny re-crawls on that schedule and only re-embeds pages whose content actually changed, so the knowledge base stays current as you update your site. You can also trigger a re-crawl manually any time.

What the agent can answer

Anything that lives in your site's page text — hours, location, services, policies, pricing, FAQs. If a visitor asks something and the agent comes up empty, check that the answer is actually on a crawled page: open the Knowledge tab and confirm the relevant page is in the crawl.

The crawler renders pages in a headless browser, so JavaScript-rendered content is generally captured. A crawl is still a snapshot, though — for data that changes by the minute (vehicle stock, appointment availability), use an integration instead of relying on the crawl.

Persona vs. knowledge

The persona(Configure → Persona) defines how the agent behaves; the knowledge base is what it knows. Keep facts out of the persona prompt — put them on your site and let the crawl pick them up, so there's one source of truth that updates itself.