AI that looks it up first, then answers with the source.

Why ordinary AI gets it wrong.

A general AI assistant answers from what it learned on the internet. It has never seen your rate card, your safety manual or last year's contract, so when it is asked about them it produces something plausible. In a business, a confident wrong figure is worse than no answer, because nobody catches it until it is a dispute.

How it works, in three steps.

• —Your documents are indexed: manuals, rate cards, contracts, procedures, past jobs. The index is built in your own accounts.
• —When someone asks a question, the system retrieves the passages that answer it, and only the ones that person is allowed to see.
• —The AI writes the answer from those passages alone, with a citation to each source. If nothing relevant is found, it says it does not know.

When your business needs it.

When the answers already exist and the problem is reaching them. The same questions go to the same two senior people all day. A new hire takes months to learn where things are written down. Staff quote from memory because finding the document takes longer than the call.

When it does not.

When the problem is missed calls, slow follow-up or work typed into two systems, RAG will not fix it, and we will say so on the call. It also cannot answer what was never written down: the first finding of many projects is the list of questions nobody has documented.

What we build and what you keep.

• —A citation on every answer, so it can be checked, not taken on trust.
• —Access controls, so people reach only what they are entitled to see.
• —An agreed refresh process for when your documents change.
• —A log of the questions it could not answer: a precise list of what your documentation is missing.
• —The index and the source documents, in your own accounts, with retention agreed before anything is loaded.