LiveKit

Build, test, deploy, and monitor realtime AI agents with flexible development tools, model integrations, and managed global infrastructure.

Build polished, brand-matched voice agent interfaces faster with reusable components, audio visualizers, and ready-made session UI blocks.

Build and optimize voice agents with access to AI models, low-latency inference, simpler model switching, and private-by-default data handling.

Connect voice agents and phone systems to callers through LiveKit numbers or third-party SIP providers.

Build livestreaming, video conferencing, and realtime applications by routing audio, video, and data between participants.

Connect, control, and monitor robot fleets across networks with low-latency transport, teleoperation, and realtime observability.

LiveKit Agent Platform

LiveKit is a developer platform for building voice, video, and physical AI agents with open-source SDKs, then deploying and monitoring them in production. Developers can customize agent logic in Python or TypeScript, integrate AI models, connect agents to telephony and external tools, and test agents against realistic scenarios. Agent Builder supports browser-based prototyping and preview, with executable Python code available for continued development. LiveKit Cloud manages deployment, dispatching, scaling, and runtime infrastructure, while observability tools provide session recordings, transcripts, traces, metrics, and logs.

Category: AI Agent Platforms

• Build Agents With Code-Level Control
  Customize agent behavior and manage agent code using software development practices such as version control, pull requests, and rollbacks.
• Scale Agent Sessions Automatically
  Adjust capacity and session placement based on demand and load, reducing the operational work of running agents at scale.
• Prototype And Preview Agents In A Browser
  Explore agent ideas and test live conversations without local setup, then deploy to LiveKit Cloud or continue development with exported Python code.
• Connect Agents To Tools And Services
  Enable agents to take action through APIs, frontend functions, data retrieval, and MCP server connections, while defining custom workflows and data collection sequences.
• Choose And Combine AI Models
  Select models and pipelines for agent workflows and change supported providers without being locked into one model vendor.
• Handle Natural Realtime Conversations
  Detect when a person has finished speaking, manage interruptions, and reduce background noise during realtime voice interactions.
• Test Agents Before Production
  Evaluate agent behavior, conversation outcomes, voice quality, and performance before deployment. Automated scenarios can run in CI and failed scenarios can block builds before release.
• Deploy And Roll Back Agents
  Ship agent updates without downtime, deploy from the CLI or dashboard, and return to a previous version when needed.
• Reach Users Across A Global Network
  Run production workloads across regions on a distributed media network designed to reduce latency and support availability.
• Inspect And Diagnose Agent Sessions
  Review session playback, transcripts, traces, metrics, and logs to understand agent behavior, investigate errors, and identify latency across the stack.
• Protect Agent And Customer Data
  Use encryption, access controls, zero-retention inference practices, and optional redaction of personally identifiable information to protect data.
• Build Multimodal AI Agents
  Create agents for voice, video, and physical AI applications using supported audio, text, transcription, video, and vision inputs and outputs.
• Turn Production Failures Into Tests
  Convert real production issues into reusable scenarios that help check for the same failures as the agent changes.
• Connect Agents To Phone Calls
  Connect agents to inbound and outbound calls through LiveKit Phone Numbers or third-party SIP providers, with support for call routing and transfers.
• Test Agent Changes Before Release
  Test agent changes in non-production deployment environments before releasing them.

Agents UI

Agents UI is an open-source component library built with React and shadcn for designing voice agent interfaces. It provides customizable audio visualizers and prebuilt UI blocks, including a control bar, chat transcript, and complete realtime session view. Start with production-ready defaults and customize the components to match your brand.

• Build Interfaces With Customizable Components
  Use reusable open-source components as a starting point for voice agent interfaces, then customize them to match a brand.
• Visualize Agent States
  Select and customize audio visualizer styles to represent states including connecting, speaking, listening, and thinking.
• Combine Core Session Elements
  Assemble a transcript, audio visualizer, and media controls into a single prebuilt view for a realtime agent session.
• Manage Calls In One Place
  Give users a single control bar for managing media, sending chat messages, and ending a call.
• Display Conversation History
  Show the agent's conversational history in a message interface, including messages and thinking status.

LiveKit Inference

LiveKit Inference provides access to more than 50 AI models for voice agents, including speech-to-text, large language, text-to-speech, realtime, and conversation-handling models. Developers can select and swap supported models through the LiveKit Agents SDK using one account and API key. The service offers low-latency routing, centralized concurrency management, inference observability, custom voice creation, and zero data retention by default.

• Access Models Through One Service
  Use models from multiple providers for voice-agent tasks through LiveKit Inference rather than integrating each supported model separately.
• Keep Inference Data Private
  Protect sensitive conversations with zero data retention by default; LiveKit and its inference providers do not log or retain inference payloads.
• Switch Models With One Line Of Code
  Try and change supported models within agent code while using one account and bill.
• Reduce End-To-End Latency
  Use routing and capacity designed to reduce the time between an agent request and its model response.
• Manage Model Concurrency Centrally
  View and manage the number of concurrent sessions connected to LiveKit Inference in one place.
• Inspect Inference Performance
  Use turn-by-turn latency statistics and request traces to understand inference performance and identify opportunities to optimize agents.
• Handle Turns And Interruptions
  Use built-in models for end-of-turn detection and adaptive interruption handling in voice-agent conversation flows.
• Create Custom Voices Across Providers
  Create a custom voice from a recording or uploaded audio and use it across supported text-to-speech providers.

LiveKit Phone Numbers

LiveKit Phone Numbers connects voice agents to the telephone network through LiveKit phone numbers and telephony infrastructure. Buy local or toll-free US numbers for inbound calls, or connect existing numbers through a third-party SIP trunk. The service supports inbound and outbound calling, routes calls between telephony systems and LiveKit rooms, and bridges phone calls into WebRTC rooms. Call handling includes IVR navigation, DTMF input, warm and cold transfers, region pinning, secure trunking, noise cancellation, and IP allowlisting. SIP signaling can use TLS and media can use SRTP when supported and configured by the upstream trunk provider.

• Choose A Local Or Toll-Free Number
  Select an available US local or toll-free number from LiveKit for inbound calls, including a preferred area code.
• Connect Phone Calls To LiveKit Rooms
  Connect telephony systems and agents with LiveKit rooms, make SIP callers available to web participants, and support live call monitoring.
• Handle Inbound And Outbound Calls
  Connect voice agents to incoming calls and configure them to place outbound calls through LiveKit telephony integrations.
• Connect Existing Phone Providers
  Bring an existing phone number from a telephony provider and connect it to a LiveKit Cloud project through a SIP trunk.
• Transfer Calls To Agents Or Numbers
  Route calls onward through cold transfers or agent-assisted warm transfers to another agent or phone number.
• Protect And Restrict SIP Connections
  Use TLS signaling and SRTP media encryption when supported, and limit which networks can initiate connections through IP allowlisting.
• Deliver Clear Call Audio
  Provide high-definition audio for both agent conversations and hold music.
• Accept Keypad Input And Navigate IVR
  Allow callers to provide DTMF input and enable agents to navigate interactive voice response systems.
• Configure Production Call Handling
  Use region pinning, secure trunking, and noise cancellation as part of configuring production phone agent calls.
• Understand Number And Call Costs
  Phone numbers have a monthly rate based on country and type, while SIP calls are billed by the minute based on direction and provider.

LiveKit media server

LiveKit's open-source media server uses WebRTC to route realtime audio, video, and data between participants. It operates as a Selective Forwarding Unit (SFU), handling media in memory for routing and any opted-in server-side processing before discarding it rather than persisting it. The server is available under the Apache 2.0 license.

• Route Realtime Media Between Participants
  Forward realtime media between participants to provide transport for WebRTC-based livestreaming, video conferencing, and other realtime applications.
• Transport Audio, Video, And Data
  Use realtime infrastructure to transport audio, video, and data between connected participants and devices.
• Inspect And Customize The Code
  Access the media server's Apache 2.0-licensed source code and use or customize it for applications.
• Forward Media Without Persisting It
  Realtime media is handled in memory for forwarding and opted-in processing, then discarded rather than persisted on LiveKit servers.
• Protect Media In Transit
  WebRTC media transport uses SRTP with DTLS-SRTP key exchange, while signaling between clients and LiveKit Cloud uses TLS 1.2 or higher.
• Keep Media Opaque To The Server
  When end-to-end encryption is enabled through the LiveKit SDK, media frames are encrypted before leaving the client so the server cannot decode their audio or video content.

LiveKit Robotics

LiveKit Robotics is a realtime platform for building and operating robotics applications. Its global edge network connects robots, operators, and AI policies for low-latency media and data transport, teleoperation, and fleet monitoring. Developers can stream video, audio, sensor data, and telemetry; send control messages; enable human-in-the-loop takeover and voice-agent control; and connect ROS topics through LiveKit Portal and LiveKit ROS Portal.

• Stream Realtime Media With Low Latency
  Transport video, audio, and data from robots and other devices over an edge network optimized for low latency.
• Stream High-Frequency Robot Data
  Send high-frequency sensor and telemetry data from robots, including information from cameras, actuators, and joint states.
• Connect Robots Across Global Networks
  Connect robots and operators across locations and network environments using LiveKit's global edge network.
• Operate Robot Fleets Remotely
  Connect robots with people and models on local networks or in the cloud to support teleoperation and fleet operations.
• Enable Human Takeover Of AI Control
  Allow AI policies to control robots while supporting human-in-the-loop takeover.
• Send One-To-One Or One-To-Many Commands
  Send and receive control messages between participants in one-to-one or one-to-many configurations.
• Control Robots Through Natural Language
  Enable voice agents to provide a natural-language interface for robot control.
• Connect ROS Topics Across The Internet
  Connect ROS-based robots and exchange ROS topics across the internet using LiveKit Portal or LiveKit ROS Portal.
• Monitor Robot Behavior And Performance
  Track robot behavior and performance across a fleet, with frame-level timestamps for media and data tracks.
• Build With Open-Source SDKs
  Use open-source SDKs for a range of languages and platforms, including C++, Rust, Python, ESP32, Node.js, React, Swift, Kotlin, and Flutter.