LiveKit Agents
Open-source framework for real-time voice and video agents on LiveKit's media servers, with phone support.
- Open source
- Developer API
A voice agent holds a spoken conversation over the phone or in an app. Under the hood it chains several systems together: speech recognition turns the caller's words into text, a language model decides what to say or which tool to call, and speech synthesis turns the reply back into audio. All of this has to happen fast enough that the pause before each answer feels natural.
The tools here split into developer platforms, where you assemble and tune each layer through an API, and finished products aimed at contact centers or small businesses, where much of that is set for you. Open-source frameworks sit at the developer end and let you run the media and agent code yourself.
Phone calls are harder than chat in specific ways. Callers interrupt, talk over the agent, use accents the speech model has not seen much, and call from noisy places. Names, addresses and account numbers are easy to mishear. Test with real recordings or real callers from your audience, and design confirmations for anything that must be exact, such as spelling back an email address.
Plan the transfer to a person carefully. The agent should know when to transfer, the person should receive a summary, and the caller should not have to repeat everything. Also decide what happens when the agent cannot reach your booking or order system during a call.
What a featured listing costs ยท from $99 a month
Open-source framework for real-time voice and video agents on LiveKit's media servers, with phone support.
Developer platform for phone and web voice agents where you choose the speech, language and voice models.
Platform for building phone agents that answer and place calls, with call flows, transfers and post-call analysis.
Phone agent platform for enterprises that handles inbound and outbound calls through scripted pathways and an API.
No-code builder for voice agents that answer calls, book appointments and connect to CRMs and calendars.
ElevenLabs platform for voice agents that use its speech models, with tools, knowledge bases and phone integration.
Voice agent platform for large contact centers, built to handle customer calls and route them to staff.
Recording and consent rules for phone calls vary by place, and some places have specific rules for automated or AI-voiced calls, especially outbound ones. Check them before you launch, and make sure the agent can tell callers they are speaking with an automated system when that is required.
Cost on voice platforms usually adds up per minute across several providers: telephony, speech recognition, the language model and the voice. Ask for a full per-minute breakdown for your expected call length, and check whether you pay for time on hold or during transfers. For high call volumes, the choice of model and voice can change running costs a lot, so test cheaper options for routine calls and keep more capable ones for harder conversations. Finally, keep recordings and transcripts of early calls, with consent, and review them often; they show where callers get stuck far better than summary dashboards do.
Ask also how the platform handles outages at any of its providers mid-call, and what the caller hears if one fails.
Modern speech models sound close to a person, but delays before replies and handling of interruptions vary a lot between platforms. Listen to full calls, not short demos, and test on real phone lines.
Yes, if you connect it to those systems. Most platforms here let the agent call your API or calendar during a call. Test what it does when that system is slow or returns an error.
Many platforms can either provide numbers or connect to an existing telephony provider. Check which carriers and systems are supported before you plan a move of your main line.
Rules differ by country and state, especially for outbound calls, call recording and automated or artificial voices. Review them with counsel for the places you call, and build disclosure and opt-out handling into the agent.