How Your Colleague Gets Its Knowledge
A managed AI Colleague doesn't answer from a single block of text. It pulls together three things on every conversation — the business details you filled in, the live reservation and listing data it reads fresh each time, and the knowledge base you give it access to.
Anatomy of an answer
Every answer is assembled from named facts.
A colleague reads one knowledge core — your business details, your live reservation and listing data pulled fresh in every conversation, and the knowledge base you author — then takes a real action. Nothing is invented, and you can see exactly which facts were used.
Diagram. Five integrations — Hostaway PMS, Airbnb, Booking.com, Stripe and HubSpot — feed a knowledge core of fourteen named facts, grouped into live PMS data, your business details, and knowledge-base articles you author. Four AI colleagues work around that core: Guest Assistant on WhatsApp and Twilio voice, Team Assistant in Slack and Microsoft Teams, Receptionist on the company phone, and Leads / Sales for inbound owners. For each question a colleague reads three or four specific facts and then emits one platform action code — ACCESS_INSTRUCTIONS, PARKING_QUESTION, WIFI_QUESTION, CREATE_TASK, CAPTURE_LEAD, UPDATE_KB, SEND_TEMPLATE or ESCALATE. The guest-facing colleague explains how a lockbox works but never reads an access code aloud.
Action codes it can trigger
Legend
Guardrail. The guest-facing colleague explains how the lockbox works and where to find it — it never reads an access code aloud. Anything it cannot ground in a named fact becomes ESCALATE: a Slack alert to ops, and a support case opened with the full transcript.
Every answer is assembled from named facts. Watch which ones a colleague reads for each question, and the action it takes as a result.
The three sources a colleague reads
When a guest sends a message or calls, the colleague assembles its answer from three sources. You control each one differently:
The structured Your details form on the colleague's Instructions tab — company name, check-in / check-out times, access instructions, and the like. These are the facts specific to your business that don't change per conversation.
Read fresh on every conversation and matched to the guest by their phone number or message thread — arrival date, what they booked, the property's details. You don't paste this in; the colleague fetches it automatically. The Live data panel on the Instructions tab shows exactly which fields it reads and what it never sees.
Your library of articles — house rules, policies, FAQs, help center content. This is the part you extend as your business grows, and where most "the colleague should know X" answers belong. You choose how much of it each colleague can read.
Giving a colleague knowledge-base access
Knowledge base access is set per colleague on the Knowledge Base tab of its detail page:
- Go to AI Agents → AI Colleagues and open the colleague you want to configure.
- Click the Knowledge Base tab.
- Choose a Knowledge base access level (see the table below).
- Optionally, scope it to specific audiences — leave audiences empty and the colleague reads articles for all audiences; select some and it only reads those (plus general articles with no audience set).
| Access level | What the colleague can read |
|---|---|
| Off | No knowledge base at all. |
| Public | Public (guest-facing) articles only. |
| Internal | Internal articles only. |
| Full hub | Both public and internal articles. |
Editing the knowledge base itself
The articles live in the Knowledge Base module (under AI Agents), not on the colleague. That separation is the point: edit an article once and every colleague with access answers with the new content — no need to touch the colleague.
You can write articles by hand, seed a batch with the setup wizard, or import from a website to pull existing content in. Mark each article public or internal so the access levels above apply cleanly.
What changed from context sources
Earlier versions let you attach ad-hoc context sources — a URL, a text snippet, or a message template — directly on the bot. That's been consolidated: knowledge now lives in one place (the Knowledge Base), and you grant access per colleague with the levels above.
- A URL you used to attach → import that page into the knowledge base instead, then grant access.
- A text snippet → make it a short knowledge base article (or, for a fact that's specific to your business, a field on the Your details form).
- A message template → still lives in Message Templates; a routing colleague can send one directly. See Colleague Triggers & Routing.