build.allanninal.dev

Part 4 of 7 · Voice agent series ~5 min read

How the brain decides what to say

The brain reads the caller’s words and decides one of four things: answer from the knowledge file, book the appointment, transfer to a human, or end the call gracefully. It’s allowed to be confident or to defer — never to invent.

The brain: four tools, one decision per caller turn A vertical flow with four branches at the bottom. At the top, the locked Utterance from the listener — the caller’s most recent words as text. An arrow leads down to the Brain, an AI with exactly four tools available. From the Brain, four labeled arrows fan out to four outcome boxes: Answer from the knowledge file (left), Book an appointment (centre-left), Transfer to a human (centre-right), and End the call gracefully (right). Each tool box describes its purpose in one short line. A bottom note reads: the AI picks one tool per caller turn. It is allowed to be confident or to defer — never to invent. Utterance caller’s words, locked by the listener Brain AI with exactly four tools, picks one Answer read from the knowledge file, say it back Book make an appointment in your calendar Transfer hand the call to a human, briefing them End close politely, log the reason, hang up The brain picks one tool per caller turn. It can be confident or defer — never invent.
Fig 4. Four tools, one pick per turn. The brain can’t answer outside this menu.

Four tools, one decision

The most important constraint on the brain is what it’s allowed to do. Most voice AIs fail because they’re given too much freedom — they invent prices, promise discounts that don’t exist, agree to refunds without checking. Here, the brain has exactly four tools available per turn, and nothing else. If it can’t fit the caller’s request into one of these four shapes, it transfers.

Tool 1 — Answer from the knowledge file

The most common tool. The brain looks up the caller’s question in the knowledge file (your hours, your services, your prices, your common FAQs), composes a short reply in your tone, and the speaker says it.

The trick is grounding: the brain answers only from the knowledge file. If the file says you close at 6, the brain says “6.” If the file doesn’t mention closing time at all, the brain doesn’t guess — it picks tool 3.

Tool 2 — Book an appointment

Optional. If your business takes appointments and you’ve connected a calendar, the brain can offer slots, take the caller’s name and number, and confirm the booking back.

The booking tool follows a strict pattern: read available times, propose, get confirmation, write to the calendar, read the confirmation back. The brain doesn’t book anything outside the offered slots, doesn’t double-book, and doesn’t make promises about cancellation policies that aren’t in the file.

Tool 3 — Transfer to a human

The escape hatch. The brain picks this when:

  • The question isn’t covered in the knowledge file.
  • The caller is upset, asking for a refund, or asking about a sensitive topic.
  • The caller explicitly asks to talk to a person.
  • The brain isn’t confident in its answer.

The transfer is short and warm: “Let me get someone who can help with that — one moment.” The brain also writes a short note for the human that picks up: what the caller wanted, what was already said, what to follow up on. The human walks into a brief, not a cold call.

Tool 4 — End the call gracefully

When the conversation has reached a natural close (caller says “thanks, that’s all I needed” or similar), the brain wraps up politely and hangs up. The audit log records what was discussed and how it ended.

What it never does

A short list of things the brain refuses to do, by design:

  • Quote a price, hour, or policy that isn’t in the knowledge file.
  • Promise anything — a refund, a callback, a delivery date — outside the booking tool.
  • Take payment information.
  • Continue answering when it’s clearly out of its depth.

If the caller pushes for any of these, the brain transfers. The shape of the four tools makes this the only natural outcome — there’s no “wing it” option in the menu.

In plain words

The brain has a small, fixed menu. Every call is a sequence of caller-turns and brain-tool-picks. The AI is fast and flexible inside that menu — it can phrase the same answer ten different ways, depending on the caller’s tone — but it never tries to do something that’s not on the menu. Boring, safe, correct.

All posts