Kiki: an ambient voice travel agent for planning trips together, powered by Vocal Bridge

An ambient agent that listens while two friends plan a Maui trip, stays quiet until it has something useful, then re-dates flights, a hotel and a car in one command when the forecast changes.

The problem: group trip planning is a conversation no tool can join

Planning a trip with other people is a messy, spoken thing. Dates, budgets, who is flying from where, who hates the rain, all of it gets negotiated between people, not typed into a form one field at a time. Then someone opens six browser tabs to turn that into a booking, and the two never stay in sync.

Every change makes it worse. Move the dates a week and the flights, hotel, rental car and activities all have to be re-checked and re-priced. A traditional flow makes one person leave the conversation, redo the search, and report back. The plan the group agreed to and the plan the tool knows about drift apart.

What they built

Kiki is an ambient voice agent that sits inside a travel app and listens while a group talks. It behaves like a calm companion, not an assistant waiting to be prompted. It tracks the trip as travelers describe it, holds each person's preferences in memory, and only interjects when it has something genuinely useful: a weather risk, a better date window, live fares to choose from.

The result is a full collaborative booking that never breaks the conversation. When Kiki flags a problem and the group agrees to move, one spoken decision re-dates and re-prices the entire trip at once, against real Sabre fares and hotels, with payment through PayPal. Voice does not replace the app; it lets the plan and the booking finally be the same thing.

How they use Vocal Bridge

Kiki is built on Vocal Bridge's Voice for your app surface and uses Vocal Bridge's Listening mode to stay present while the travelers talk to each other, without demanding a prompt or taking every turn. The voice agent and the UI share one real-time data channel. As travelers mention Maui, five people, and American Airlines out of San Francisco, the sidebar updates live. When Kiki runs a Sabre search, the UI moves from searching to booking and surfaces the options Kiki is describing.

The split is clean. Kiki's team owns the travel logic: Sabre integration, preference memory, the PayPal step, and the living itinerary. Vocal Bridge owns the voice channel: streaming speech, turn detection and barge-in, and the live transcript.

import { useAgentActions } from "@vocalbridgeai/react";

const { onAction, sendAction } = useAgentActions();

// agent → app: Kiki found live fares, render them
onAction("present_flights", ({ options }) => {
  setItinerary((it) => ({ ...it, flights: options, stage: "booking" }));
});

// app → agent: a traveler picked a card on screen
const handleSelectFlight = (id) => sendAction("select_flight", { id });

Illustrative. The production Kiki handles richer itinerary state across flights, stay, transport and activities.

What's hard about voice here

• Knowing when to stay quiet. Kiki listens to two people talking to each other, not to it, and has to tell "still deciding" apart from "now is the moment to help." Vocal Bridge's Listening mode keeps Kiki present without forcing the travelers to address it, while turn detection and barge-in let it hold its turn and yield instantly when a person speaks.
• Dead air while Sabre runs. Live lookups take real time, and silence reads as broken. Kiki narrates as it works, which Vocal Bridge's streaming speech makes land immediately rather than after the tool call finishes.
• Keeping plan and screen in sync. One "move it to August" has to re-date flights, hotel and car and reflow the itinerary, and the words and UI cannot disagree. The bidirectional data channel keeps the spoken plan and the on-screen plan on the same state.

Kiki's demo establishes a genuinely different interaction pattern. The production UX levers follow directly from that strength: make the listening state unmistakable, show who and what Kiki is remembering, and give every traveler an obvious way to pause or correct the agent.