The AI assistant & knowledge base
In one sentence
Feed MsgBuddy your own documents (your knowledge base), and the AI assistant can answer customers’ questions in natural language — using your information, not made-up facts.
How it works, simply
Unlike a flow (which follows a script) or a rule (which matches keywords), the AI understands a question and writes an answer. To keep it accurate, it only draws on the documents you give it.
This technique — answering from your documents — is why the assistant can say “we’re open till 8 on weekends” only if that’s actually in your knowledge base.
Setting up the knowledge base
- Go to Settings → Knowledge.
- Add documents: FAQs, policies, your menu, shipping info, returns — anything customers ask about.
- MsgBuddy processes each doc so the AI can search it when answering.
Write knowledge docs the way a customer asks, not the way a lawyer writes. Short Q&A snippets (“Do you deliver to Navi Mumbai? Yes, ₹40, same day if ordered before 4 p.m.”) work far better than a dense wall of text.
Turning the assistant on
In Settings → Chatbot you choose how it runs:
| Option | What it means | Who it’s for |
|---|---|---|
| Off | No AI replies | You prefer rules/flows/humans only |
| Use my own key | You bring an API key from Anthropic or Google | You want to use and pay your own AI provider |
| Use MsgBuddy AI | Managed for you — nothing to configure | You just want it to work; usage is metered |
“Bring your own key” vs “MsgBuddy AI”: with your own key, the AI usage bills through your provider account. With MsgBuddy AI, it’s handled for you and counted under your Usage. Same assistant either way — just a different engine behind it.
Keeping it safe and on-brand
- Give it a system prompt (a short instruction like “You are Sunrise Bakery’s friendly assistant. Only answer about our products, hours, and delivery. If unsure, offer to connect a human.”).
- The AI only answers when it’s the one driving the conversation — and a human can take over instantly.
AI is powerful but not infallible. For anything high-stakes (payments, complaints, legal), design your setup to hand off to a human. Keep the knowledge base accurate — the assistant is only as good as what you feed it.
How this connects
- The assistant is one responder in Who answers?.
- Its usage shows up in Usage (especially on MsgBuddy AI).
- A human claiming the chat silences it — the Inbox control model.
Next: the rule that ties it together — Who answers? →.