# Teach it about your database Every database has quirks the schema does not spell out: a table named nothing like what it holds, a column that really means "soft deleted", a rule like "an active customer is one with an order in the last 90 days". Without help, you end up pasting the same explanations into every chat. SnoutData keeps a **knowledge base** of facts about your database. Teach it a fact once and it applies to every future question on that connection, so you stop repeating yourself. ## The motivating example > "For orders, use the `checkout_session` table and its `completed_at` field. There is no > `orders` table, and the columns are not commented." Tell the assistant that once, and from then on a question like "show me last month's orders" resolves to `checkout_session.completed_at` instead of guessing at a table that does not exist. ## It offers to remember As the assistant writes or corrects SQL, it notices reusable facts and proposes them. A gentle, non-blocking card appears in the chat: > 💡 **Want me to remember this?** > "There's no orders table, orders live in `checkout_session` (`completed_at` is the order > date)." > **[ Remember ] [ Not now ]** - **Remember** makes the fact active. It is now included with every future question on this connection. - **Not now** does not discard it. It drops into the list below as a greyed "suggested" item you can keep later. Nothing is ever saved without your confirmation. The assistant proposes; you decide. ## See what it knows A chip in the chat header shows how many facts the assistant has for the current connection: > 🧠 **Knows 4 things about this DB** ▾ Open it for a plain-language list, scoped to the connection: - Toggle a fact on or off without deleting it. - Edit the wording, or delete it. - Accept or dismiss a suggested fact. - Use **Teach it something** to add your own fact by typing a sentence. There is no config file, no rules syntax, and no scope dropdown to learn. You write a sentence and the app works out where it applies. ## Add your own facts Two ways to add a fact directly: - The **Teach it something** box in the in-chat list. - **Settings → Knowledge**, a fuller manager where you can review, edit, and organize facts for any connection, not only the one you are connected to. By default a fact you type applies to the current connection. An optional "applies to every database" toggle promotes it to a **general rule** that applies everywhere, which is the right home for preferences like "always use CTEs". General rules are also available in Settings → Knowledge even when no connection is selected. ## How facts are used Active facts are included in the assistant's context alongside your schema, so the model reads them as authoritative when it writes a query. Because the important facts are always included rather than searched for, a typo in your question ("custumer sales") cannot cause the assistant to miss them. Knowledge is stored **on your machine**, next to your connections. It is not a shared cloud index. Like your schema, the facts are sent to the model as part of the prompt when you ask a question. ## Privacy and control - Every fact is either typed by you or confirmed with a single tap. The app never learns silently. - You can see, edit, toggle off, or delete anything the assistant knows. - Facts are per-connection by default, so a fact about one database does not leak into another.