Knowledge & behaviour
Train your agent
Websites, sitemaps, single pages, files, text, Notion and Shopify products — and how to keep answers up to date.
The agent only answers from what you give it. When something isn't in its knowledge it says so and offers your team, instead of guessing.
Add knowledge
Open Build → Data sources and choose a type:
| Type | Use it for | Notes |
|---|---|---|
| Website | Your whole site | Give your homepage; the agent follows links. Limited by your plan's pages per crawl. |
| Sitemap | The most complete crawl | Give https://yoursite.com/sitemap.xml. |
| Single page | One URL only | For example a price list or a policy page. |
| Files | Documents | PDF, DOCX, TXT, MD, CSV and HTML, up to 25 MB each. |
| Text | Anything you paste | Policies, price lists, opening hours. Markdown headings (#) help the agent. |
| Notion | Pages and databases | See Notion. |
| Shopify | Your product catalogue | Imports products with prices, variants and stock. See Shopify. |
Each source shows its progress while it's learning. You can keep chatting in the Playground meanwhile.
Focus a website crawl
Open Advanced filters when adding a website:
- Only include URLs containing — e.g.
/docs, /help(comma separated). - Exclude URLs containing — e.g.
/blog, /careers.
Keep the crawl to pages customers would ask about. Fewer, better pages give sharper answers and stay inside your plan's training limit.
Q&A always wins
Build → Q&A holds exact answers for exact questions. They take priority over everything else, so use them for prices, offers and anything that must be word-perfect. See Q&A.
Business overview
After training, the agent writes a short overview of your business — what you sell (with prices), how to buy, delivery, policies and contact details. It uses this to answer general questions such as What do you sell?. Check it under Build → Data sources and edit it if something is off.
Keeping it up to date
- Click Retrain on a source whenever your site or document changes.
- Turn on Settings → AI & answers → Auto-retrain websites daily (higher plans): website sources are re-crawled every night and only changed pages are re-learned. Shopify products re-sync daily too.
- Activity → Needs review lists questions the agent couldn't answer well, so you know what to add. See Needs review.
Training limits
Your plan sets how much text all sources of an agent may hold (in MB) and how many pages one crawl reads. Training data used on the Data sources page shows where you are. Remove sources you no longer need to free space.
Add knowledge from code
Use POST /api/v1/agents/{id}/sources to add text, pages, sites, sitemaps or Shopify stores, and POST /api/v1/agents/{id}/qa for Q&A. See the Sources API.