Soot answers customers from your own website
Prices and policies change; a pasted copy in a prompt goes stale. Soot looks facts up from your own web pages, records what it cannot answer, and tells you when several customers hit the same gap.
If you run a small shop, you have probably done this: pasted your price list and return policy into the instructions of an AI helper. A week later the price changes on your website. The pasted copy does not. A customer asks, and the helper quotes the old price.
Soot avoids this by not keeping a copy. It looks facts up from the pages you have already published. The Soot data docs put the rule this way: project facts flow from the project, not from the prompt.
What Soot does, step by step
The example below is a small pottery shop, made up to keep things concrete.
1. You write the facts on your website, one page per topic. One page for shipping, one for returns, one for caring for glazed ceramics. Prices should come from the shop's own catalog, not be typed into the page. The Soot docs are blunt about it: a page with a hard-coded price has brought the stale copy back.
2. The shop publishes a `/llms.txt` file. It is an index page in the llms.txt format: a title, a one-line summary and links to each page. This step and the next are a bit technical. The Soot docs write the setup part for a coding agent to do on your behalf, so you can hand it to whoever builds your site, or to a coding agent. In Nest, Settings → Public knowledge has a "Copy prompt for assistant" button for exactly that.
3. The customer-facing Soot gets lookup turned on. This is declared in configuration, and Nest has a switch for which pools may hold knowledge. Installing or selecting a pack does not grant access by itself (the post Installing a Forge pack does not mean a Soot may use it covers that). For knowledge, the Soot can only read the pools you listed.
4. The runtime syncs by itself. Every few minutes it checks which sources are due, downloads them and splits each page at its ## headings into chunks that make sense alone. If a download fails, the old copy stays. The docs' reasoning: dated old facts are better than none.
5. A customer asks, Soot looks it up and answers with the source. The answer cites the page it read. The Soot's own memory and its earlier answers do not count as sources.
When your website has no answer yet
Say a customer asks, "Can this mug go in the microwave?" and the shop never wrote a line about it. Soot does not guess. It says it cannot confirm that, and records the gap with a dedicated call (knowledge_gap) so someone can fill it in.
One customer asking is just listed. If the same gap is asked in at least 2 different conversations within 14 days, it becomes an item in Cần bạn ("Needs you"), the work inbox in Nest. You add a paragraph to the website and press Đã bổ sung ("Added"). The runtime syncs that source right away.
To check that the new page is found, open Soot tìm thấy gì? ("What does Soot find?") in Nest and type the words a customer would use. It runs exactly the lookup Soot uses. Search is keyword-based, so write your pages in the words customers use: if they ask about the microwave and your page only says "oven use", it may not match.
If the question keeps coming back, the item reopens in Cần bạn on its own: the fix did not work, or the page has not synced yet.
Before you try it
- Keep private data out of a knowledge pool. A public Soot reads only what the project has published, and that is all this pool should hold.
- A Soot that talks to customers can use knowledge lookup, but in the public setting it cannot have a data pool or other tools attached.
Next step
Pick the three questions your customers ask most and check whether your website answers each one. Write a page for any it does not. The mechanism is in the data and knowledge docs, and you can get Nest at trysoot.com/download.