A real-time voice receptionist, built around everyday calls.
Koala AI built a working prototype and tested it with one dental clinic. This page explains what it can do, the technology involved and the questions that remain before a wider rollout.
This is a working prototype, not a wider production rollout. No measured business results are claimed.
The problem we explored
Appointment requests and routine questions can compete with the people at the front desk. We explored whether a voice agent could handle defined conversations while keeping the clinic’s rules and human fallback central to the design.
What the prototype does
- Understands spoken requests through telephone and web-based voice interactions.
- Answers configured questions, such as opening hours and location.
- Checks appointment availability in its configured workflow.
- Handles booking, cancellation and rescheduling requests.
These capabilities describe the tested prototype. They do not mean every clinic’s scheduling software or phone system is already supported.
How the configured workflow works
- Caller asks about an appointment
- Agent checks the configured workflow
- Agent responds within agreed rules
- Human fallback needs implementation testing
The technology behind it
The prototype combines the OpenAI Realtime API, LiveKit, Next.js and AWS. These bring together real-time voice interaction, the application experience and hosting infrastructure.
A useful voice agent needs more than a natural conversation. Its actions must be tied to agreed business rules and the connections available in the target systems. Each new implementation needs its own access checks, workflow configuration and testing.
Testing and evidence
Testing with one dental clinic covered the capabilities described above. Published evidence does not include call volumes, accuracy scores or time savings.
A demonstration can show the current behaviour. A pilot for another business would need agreed acceptance checks and a review of its own systems.
Limits that matter
- No approved or certified practice-management integrations are claimed.
- Live call transfer depends on the phone system and configuration.
- After-hours coverage and fallback arrangements must be agreed and tested before launch.
- The agent does not provide clinical advice or emergency triage.
- Data locations, access, retention and provider terms must be reviewed before connecting client or patient information.
What a next pilot would establish
We would first map the intended calls and actions, confirm system access, and agree how uncertain requests return to a person. Testing would use suitable sample data before client data is connected. Scope, acceptance checks and launch readiness would be agreed with the business.
Explore a similar workflow
Tell us about the calls your business handles and the systems involved. We can assess whether a defined pilot is a practical next step.
Arrange a 15-minute introduction