Build Your Knowledge Base From Your Inbox

The AI reads your real inbox and turns your team's own answers into reusable articles — you approve every one.

Why this works

A knowledge base is what turns a generic colleague into your colleague: it's the set of facts your AI Colleague searches to answer a guest. The hard part is usually writing it. But you've already written most of it — one message at a time — every time your team answered a guest.

Your check-in instructions, parking notes, WiFi details, and the hundred small clarifications your team sends are the source of truth. This flow simply reads them back and distills them into short articles your colleague can reuse — so the same question never needs a human answer twice.

Before you start

• Connect your AI over MCP. A workspace admin generates a token and links Claude to your workspace — see Connect Your Own AI (MCP). The token scopes everything to your workspace only.
• Have your inbox synced. Connect your PMS so guest conversations flow into your workspace — see Connecting Enhanced Rentals to Hostaway. The more history, the better the analysis.
• Know which colleague answers guests. This is the AI Colleague whose knowledge base you're filling — usually your main Guest Relations colleague.

How it works

1. 1

   It reads your synced inbox

   Your AI pulls your recent guest conversations — both what guests asked and how your team replied — plus your colleague's current prompt and any existing knowledge base articles. Nothing is uploaded anywhere; it reads through your workspace over the connection.

2. 2

   It analyzes what guests actually ask

   Instead of guessing, it ranks the real questions by how often they come up — check-in and access, parking, WiFi, pets, cancellations, what's nearby. That ranking becomes your article backlog, most valuable first.

3. 3

   It drafts the articles — with you

   For each top question it drafts a short, factual article, grounded in how your team already answers (your welcome messages, check-in instructions, past replies). You review, correct anything, and decide what goes in. Nothing is invented — if a fact isn't in your data, the draft defers instead of guessing.

4. 4

   It adds them to your knowledge base

   Once you approve, it writes the articles straight into your knowledge base. If it also improves your colleague's prompt, that change is saved as a draft you publish — your live colleague is untouched until you say so.

What to say to your AI

Once connected, just ask in plain language. A natural sequence:

• “Read my last 100 guest conversations and tell me the top questions guests ask.”
• “For each of those top questions, draft a short knowledge base article using how my team already answers them.”
• “Add these articles to my knowledge base, audience: guest.”
• “Check my AI Colleague's prompt against these — anything missing or contradicting? Propose an updated prompt as a draft.”
• “What questions come up that we don't have a good answer for yet?”

Work through the drafts together — tighten wording, fix a fact, drop anything you don't want public — then let it add the ones you approve.

Turn the knowledge base on

One step is done by hand, on purpose: open your guest colleague and switch Knowledge base access on (set it to your guest audience). Until you do, the articles exist but the colleague won't read them. See Bot Context Sources for where this lives.

Is it safe?

• Your data only. The token maps to your workspace and can never reach another company's data.
• You approve every change. The AI drafts; you decide what gets added. Articles are easy to edit or delete, and any prompt update is saved as a draft you publish yourself.
• Nothing invented. Articles are grounded in your own data. When a fact isn't known, the draft says so and points the guest to the right place instead of guessing.
• It's your AI. The conversation happens in your own Claude — we don't see it.