Knowledge
The knowledge base is the information your agent can draw on when answering questions. You can add sources as crawled websites, individual URLs, or manually written question-and-answer pairs. Elaras indexes the content and retrieves the most relevant chunks at inference time.
Source types
| Type | Best for |
|---|---|
| URL | A single page — FAQ, pricing page, terms, product documentation |
| Crawl | An entire website or section of a site — Elaras follows internal links up to the depth you set |
| Q&A | Hand-crafted answers to specific questions the agent should always answer a particular way |
Adding a URL source
In the Elaras dashboard, open your agent and go to Knowledge > Add source > URL.
Paste the URL and click Add. Elaras fetches the page, strips navigation and boilerplate, and indexes the content. Indexing usually completes within a few seconds for a single page.
Adding a crawl source
Go to Knowledge > Add source > Website crawl. Enter the start URL and configure the crawl depth (how many links deep to follow from the starting page). A depth of 1 only indexes the start page; 2 follows links on the start page; and so on.
Large crawls run in the background — the dashboard shows indexing progress. You can start using the agent while the crawl completes; the knowledge base updates incrementally.
Adding Q&A pairs
Go to Knowledge > Add source > Q&A. Enter a question and the exact answer you want the agent to give. Use Q&A for answers that need to be precise — pricing, legal disclaimers, support contact details — where you do not want the agent to paraphrase.
Resyncing sources
URL and crawl sources are not automatically updated when your website changes. To pull in the latest content, resync the source from the dashboard (Knowledge > source menu > Resync).
A resync re-fetches the original URL or re-crawls the site and re-indexes the content. The previous version remains active until indexing completes.
Deleting sources
Select a source from the Knowledge list and click Delete. This removes its indexed content immediately — the agent will no longer draw on that source for new conversations.
Tips
- Be specific with URLs. Adding the exact page that answers a question (e.g.
/pricing) is more effective than crawling the whole site if only that page is relevant. - Use Q&A for invariants. Facts that must never be paraphrased — prices, legal copy, contact details — should be Q&A pairs, not crawled pages.
- Crawl depth vs. breadth. A depth of 2–3 covers most documentation sites. Very large crawls slow indexing and can include irrelevant pages like blog archives.
- Resync on publish. Add a resync call to your deployment pipeline so the knowledge base stays current whenever your website is updated.