LiveKit Agents

LiveKit Agents is a developer framework and cloud platform for building real-time AI agents that can join live voice, video, telephony, and multimodal conversations, with a free Build plan at $0/month and paid Ship and Scale tiers starting at $50/month and $500/month minimums. It is best understood as programmable real-time media infrastructure with an agent runtime layered on top: developers write agent workers, connect speech and language models through plugins or LiveKit Inference, and then deploy those agents into LiveKit rooms, WebRTC applications, or SIP phone workflows. That makes it a strong fit for teams that need more control than a turnkey phone-agent builder provides, especially when the product involves video, browser or mobile SDKs, custom frontend state, real-time interruption handling, or self-hosted media infrastructure.

The public LiveKit materials give several concrete facts that help define the product. The Build plan is $0/month and includes 1,000 free agent session minutes monthly, 1 free LiveKit phone number, agent deployment, agent observability, inference credits, the global edge network, and session metrics and analytics. The Ship plan has a published minimum of $50/month and adds team collaboration, rollback to previous agent deployments, email support, and shared production billing across projects. The Scale plan has a published minimum of $500/month and adds role-based access, metrics export APIs, region pinning, security reports / HIPAA, and inference discounts. LiveKit's pricing calculator lists agent session usage at $0.0100 per minute and telephony usage at $0.0100 per minute before model-specific inference charges. LiveKit's quotas documentation describes the free Build quota as 1,000 agent session minutes, 100,000 agent observability events, 1,000 minutes of agent audio recordings, $2.50 in LiveKit Inference credits, 50 US local inbound minutes, 1 included US local phone number, and 1,000 third-party SIP minutes.

For engineering teams, the main advantage is architectural flexibility. LiveKit Agents can use the common STT-to-LLM-to-TTS pattern, but it also supports realtime speech-to-speech model integrations where available. The AgentSession abstraction coordinates user input, model calls, tool use, output speech, observability events, turn-taking, and conversation state, while LiveKit rooms handle media tracks and participant connectivity. Teams can connect an agent to a web or mobile client through LiveKit SDKs, expose it to the phone network through SIP, or place it inside a broader video or multimodal application where the agent needs access to audio, video, data messages, or application-specific state.

The tradeoff is that LiveKit Agents is not a no-code product. Teams should expect to build and operate Python or Node.js agent code, choose and configure model providers, manage tool-calling behavior, test latency under realistic load, and monitor usage across plan fees, agent session minutes, telephony, and inference. Security-sensitive teams should also distinguish between LiveKit Cloud features and self-hosted responsibilities. LiveKit publishes SOC 2 Type II, GDPR, HIPAA BAA availability for eligible customers, encrypted WebRTC media transport, JWT-based room access, and open-source components, but deployment-specific controls such as retention, observability access, recording storage, and compliance evidence still need to be validated against the team's architecture and contract.