Real-time media
LiveKit manages SIP calls, media transport, rooms, and agent dispatch.
AI voice agent
A voice agent that can speak and use business tools
An AI voice agent is software that receives telephone audio in real time, interprets the caller’s current goal, selects an allowed action, uses a business tool, and returns a spoken response. OMO separates probabilistic language interpretation from critical business execution so outcomes can be validated, repeated safely, and audited.Who is this for?
Teams moving a repeatable phone workflow into controlled automation while retaining clear human escalation paths.
What problems does it solve?
- Moving chatbot logic directly into telephony without reliable turn-taking
- Confusing a generated answer with a completed business action
- Evaluating call quality without audio, events, and latency evidence
What can OMO do?
A practical system, not an isolated model.
Stateful conversation
Pipecat flows define stages, allowed transitions, and the answer expected at each point.
Deterministic tools
Calendar, lead, survey, and routing actions run through typed server-side services.
Complete telemetry
Transcript, events, latency, and the final outcome are joined into one session.
How it works
From the caller’s first sentence to a verified result.
- 01
Listen
The audio pipeline can receive the caller while the agent is speaking.
- 02
Interpret
STT and the language model evaluate the answer against current state and confirmed facts.
- 03
Execute
The selected operation is passed to a validated business service.
- 04
Respond
TTS produces a concise response based on the actual tool result.
Evidence retained
Every completed action has an operational record.
- Flow version and current stage
- Tool request and returned result
- Transcript, audio, and latency breakdown
System boundaries
What OMO does not assume or promise.
- The LLM is not the source of truth for calendar or database state.
- The agent can use only explicitly approved tools.
- Irreversible actions require defined validation or confirmation.
Direct answers
Questions businesses ask before implementation.
What components make up an AI voice agent?
The core chain is telephony → audio processing → STT → state and conversation control → business tool → TTS. A production system also needs validation, telemetry, failure handling, and human escalation.
How is a voice agent different from IVR?
Traditional IVR follows fixed keypad menus. A voice agent can interpret natural speech, but a reliable system still uses controlled states and approved actions rather than unrestricted generation.
Can the agent hear a caller while it is speaking?
Yes. With properly configured full-duplex audio and turn-taking, incoming speech is captured during playback. A separate policy decides whether the utterance should interrupt, be buffered, or be ignored.