Skip to content

Use cases

Six places a voice agent pays for itself

Every one of these leans on something the platform does today — tool calling, barge-in, threads, transcripts — rather than on a demo that only works in a quiet room.

Customer support that resolves, not routes

Answer account questions on the phone or in-app, and hand off with context when it matters.

The agent looks up the caller's real account through your own HTTP tools rather than guessing from training data, so it can answer 'where is my order' with an actual tracking number. When it cannot resolve something, the transcript and the structured context travel to the human who picks it up.

Illustrative exchange

Caller: Hi — where's my order?

Agent: Order 4471 shipped Tuesday and is out for delivery today. Want the tracking link?

Caller: Actually, can I change the address—

  • Tool calling for order, account and ticket lookups
  • userInfo and metadata put plan and entitlement context in the prompt
  • Full transcripts persisted per turn for handoff and audit
  • Barge-in, so a caller can interrupt a long answer

Voice ordering and reservations

Take an order over the phone, confirm it against live inventory, and write it into your system.

Menus, availability and prices come from your endpoints at call time, so the agent never quotes a sold-out item. Because it is a tool call and not a prompt, changing the menu does not mean re-writing the agent.

Illustrative exchange

Caller: Two large pepperoni, one with no cheese.

Agent: Two large pepperoni, one without cheese — that's $27.40 for pickup at 6:15.

Caller: Make it 6:45.

  • Tool calling against live inventory and pricing
  • Threads keep a repeat caller's context across sessions
  • Barge-in for the mid-order change that always comes
  • Transcripts as an order record

Scheduling and reminders

Book, move and confirm appointments against a real calendar, without a phone tree.

The agent reads live availability and writes the booking back through your tools, so there is no reconciliation step afterwards. Callers reschedule in one turn instead of navigating a menu.

Illustrative exchange

Caller: I need to move Thursday's appointment.

Agent: I have Friday 10:00 or Monday 14:30. Which suits?

Caller: Friday works.

  • Tool calling to read availability and write bookings
  • userInfo carries the caller's identity into the prompt
  • Threads so a callback continues the same conversation

An in-app voice assistant

Add a talk button to an app you already ship, on every platform at once.

One agent configuration drives Web, React Native, iOS, Android and Flutter over the same wire protocol, so the assistant behaves the same on a phone as in a browser. The client authenticates with a publishable key, which is safe to ship inside a bundle.

Illustrative exchange

Caller: Show me last month's invoices.

Agent: Opening March — eleven invoices, two unpaid. Want the unpaid ones only?

Caller: Yes, and export them.

  • Five client SDKs on one protocol
  • Publishable keys with an origin allowlist for shipped clients
  • Session tokens minted server-side for stricter deployments
  • Tool calling to drive the app itself

Hands-free field and warehouse work

Log work, look up procedures and capture readings while both hands stay busy.

Technicians, drivers and warehouse staff cannot stop to type. The agent captures structured data through tool calls as they talk, and the transcript stands as the record of what was done.

Illustrative exchange

Caller: Unit 12, pressure reads 4.2 bar.

Agent: Logged 4.2 bar on unit 12 — that's below the 4.5 threshold. Raise a ticket?

Caller: Raise it, priority high.

  • Tool calling to write structured records mid-conversation
  • Barge-in and interruption for noisy, interrupted environments
  • Native iOS and Android SDKs for devices in the field
  • Transcripts persisted per turn

Tutoring and conversation practice

Low-latency back-and-forth where interruption is the point, not an edge case.

Practice only works if the learner can interrupt, be interrupted, and hear a correction immediately. Barge-in is built into the pipeline: when the learner speaks, playback stops and the partial response is discarded rather than queued.

Illustrative exchange

Agent: ¿Cómo fue tu fin de semana?

Caller: Yo… fui a la playa con mi—

Agent: Bien. 'Fui a la playa' — perfect. Keep going.

  • Barge-in with the partial response discarded, not buffered
  • Threads so progress carries across sessions
  • additionalSystemPrompts to set the level and correction style
  • Transcripts for review after the session

Not on this list?

The pipeline is the product, not the script

None of these are separate products or templates. They are the same agent configuration — a prompt, a set of HTTP tools, and a voice — pointed at a different problem.

Screening and intakeOutbound follow-upAccessibility interfacesKiosks and drive-thruInternal helpdeskVoice-driven search

Hear it before you build on it

The demo is the real pipeline, not a recording. Talk to it, interrupt it, and decide from there.