LiveKit

What it is for

LiveKit is an Apache-2.0 realtime stack: put humans and agents in a WebRTC room, then optionally host the agent on LiveKit Cloud. Use it when the job is a voice or video agent with rooms, phones, and traces, not when you need a hosted STT API with no media server.

What makes it unique: the cover is the official livekit/livekit GitHub file list — pkg/, cmd/, deploy/, Apache-2.0 in the About pane — not a pricing card.

Key features

• Voice Agents — lk agent init then a Python/Node AgentSession (STT→LLM→TTS or a realtime model) you can talk to in the terminal or Console (Voice AI quickstart; Agents Quickstart video). One session; those model names are SKU. Agent Builder is a shorter path onto this, not a second product.
• Rooms / WebRTC — the media server that connects browsers, native apps, and agents under 250 ms (livekit.io; pricing Participants). Transport, not the LLM.
• Cloud deploy — lk agent create hosts the same agent (Build 1 deployment / 5 concurrent sessions; Ship 2 / 20; Scale 4 / up to 600) (pricing). A host, not the framework.
• Telephony — US local / toll-free numbers, SIP, Twilio, WhatsApp into the same room (pricing Telephony). A phone pipe, not a WebRTC minute.
• Observability — session recordings, turn traces, text/audio simulations (pricing Testing & observability). An ops desk, not a generate.

Self-host (livekit-server + Agents) is the same stack you run. lk / SDKs are install. Agent-session minutes, inference credits, and WebRTC minutes are the meter.

Who it's for

Best fit: voice teams who will burn Build for $0 (1,000 agent minutes, $2.50 inference credits, 1 free number) and only open Ship at $50 when they need a second deploy and email support.

Pricing

Verified 2026-10-22 (Asia/Shanghai) from livekit.io/pricing and the Apache-2.0 LICENSE. Cloud cards plus usage. Software is free to self-host.

• Software (OSS) — Apache 2.0. License fee $0. You host the media server and the agent process.
• Build — $0/month. No card. 1,000 agent session minutes; 5 concurrent sessions; 1 deployment; $2.50 inference credits; 5,000 WebRTC minutes; 1 free US local number + 50 inbound minutes; 60 shared transcode minutes.
• Ship — starting at $50/month. 5,000 agent minutes then $0.01/min; 20 concurrent; 2 deployments; $5 credits; 150,000 WebRTC minutes then $0.0005/min; team collab; email support.
• Scale — starting at $500/month. 50,000 agent minutes then $0.01/min; up to 600 concurrent; 4 deployments; $50 credits (discounted inference); RBAC; region pinning; HIPAA / security reports.
• Enterprise — custom. Volume, SSO, Slack, SLA.

Worked estimator on the page: ~$0.0479/min for a sample LLM+STT+TTS+observability stack — not a plan card. Confirm checkout; overage is metered.

Limitations / costs

• Build $0 is not free tokens. Inference credits and session minutes still cap out.
• Ship / Scale floors are $50 / $500. Usage on top.
• A room + agent host, not an STT vendor. For hours of files only, see Gladia.

Alternatives

• Vapi — another hosted voice-agent desk when you do not want to run rooms yourself.
• Gladia — speech-to-text API when you only need a transcript, not a room.
• ElevenLabs — TTS / voice agents when LiveKit’s room stack is not the job.

Official tutorial

Official LiveKit channel video: LiveKit Agents Quickstart: Create your first Agent (oEmbed author: LiveKit).

Written path: Voice AI quickstart plus Pricing. Channel also has Build a Voice Agent in Minutes (same official author).

Website

https://livekit.io

Repository

https://github.com/livekit/livekit

Source

Catalog row curated for AI Tools Nav. Cover is the official livekit/livekit GitHub file list (Apache-2.0 + pkg/cmd). Dollars checked on livekit.io/pricing (Build $0; Ship $50; Scale $500; Enterprise custom). Tutorial oEmbed author: LiveKit.

Last verified

2026-10-22 (Asia/Shanghai) editorial check.