Does the AI need training on our product?+
No training in the classical ML sense. Upload your docs (PDFs, Markdown, DOCX, plain text) to the knowledge base and the AI retrieves the relevant chunks on every conversation turn. Changes to your docs are live in seconds — re-upload the file and the next conversation uses the new content.
How does it handle questions about pricing?+
Two ways. (1) Upload your pricing page or pricing PDF to the knowledge base — the AI cites it on every pricing question. (2) Enable web search and the AI fetches your /pricing page live, so even price changes you forget to re-upload still flow through. Most SaaS teams use both.
What if the AI gets a feature question wrong?+
Two layers of safety. The AI is instructed to only answer from your uploaded content — if the knowledge base does not cover a feature, it asks a clarifying question or hands off to a human rather than guessing. If it still gets something wrong, you tweak the docs (or add a more specific entry) and the correction is live on the next conversation.
Can it deflect bug reports vs sales questions?+
Yes. The AI detects when a conversation is about an async issue (bug, refund, account problem) and files a ticket automatically, attaching the full transcript. Sales / pricing / feature questions stay in the conversation flow — the AI answers and offers to book a demo when relevant.
How does Agentbot handle visitors who insist on talking to a human?+
The AI escalates: it asks one clarifying question to gather context, then flags the conversation as needs_human. The chat doesn't go silent — the AI stays available to help while the team picks it up. Your inbox shows an 'Unassigned' alert with a sound notification.
Will the AI work for a 5-person SaaS or only at scale?+
It's specifically built for the 1–20 person team — no agent cap during beta, no per-resolution charges, no per-seat tax. The AI handles the bulk of inbound chat so a small team can ship product instead of triaging support.