LiveKit is the realtime transport layer under a large share of 2026's voice agents: it moves audio between a user and your agent with low enough latency that interruption handling works. The pricing is not one meter but several stacked on the same call — participant minutes, agent minutes, and egress — which is why estimates built on a single rate come out wrong.
| Tier | Price | Included | Overage & notes |
|---|---|---|---|
| Build | Free forever | 5,000 WebRTC min, 1,000 agent min | +50 GB egress, $2.50 inference credit |
| Ship | $50 / mo | 150,000 WebRTC min | Then ≈$0.0005 / participant min |
| Scale | $500 / mo | 1.5M WebRTC min | Then ≈$0.0004 / participant min |
| AI agent session | ≈$0.01 / min | Metered separately | Usually the dominant meter |
→ Estimate your bill on the API cost calculator or model a whole app with the AI app cost estimator.
Yes — the Build tier is permanently free, not a trial, and includes 5,000 WebRTC minutes, 50 GB egress, 1,000 AI agent minutes and $2.50 of LiveKit Inference credit every month. A thousand agent minutes is roughly sixteen hours of conversation, which covers development and a small pilot outright. See how that stacks up on the free API tiers page.
1. Sign up at cloud.livekit.io and create a project.
2. Copy the API key, secret and project URL from project settings.
3. Mint access tokens server-side with the key and secret — never ship the secret to a client.
4. Build the agent with the Agents SDK; the Build tier covers development without billing setup.
Test it:
The agent meter, not the transport. At roughly $0.01 per agent minute against $0.0005 per participant minute, agent time costs about twenty times what moving the audio costs — so an estimate built from the headline WebRTC rate will land an order of magnitude low. And none of it includes the models: speech-to-text, the LLM turn and text-to-speech are billed by whoever provides them. Price the full stack with the AI voice agent cost calculator and the realtime voice API cost calculator, and compare speech providers via Deepgram, AssemblyAI and ElevenLabs.
Reference pricing (July 2026): four tiers — Build (free), Ship at $50 a month with 150,000 WebRTC minutes included, Scale at $500 a month with 1.5 million minutes, and custom Enterprise. Past the included allowance, WebRTC participant minutes run about $0.0005 each on Ship and $0.0004 on Scale. AI agent sessions are metered separately at roughly $0.01 per agent minute, and egress is billed by the gigabyte.
Yes, and it is permanent rather than a trial. The Build tier includes 5,000 WebRTC minutes, 50 GB of egress, 1,000 AI agent minutes and $2.50 of LiveKit Inference credit every month. For a voice agent in development that is roughly sixteen hours of agent conversation a month at no cost.
Because a voice call bills on more than one meter at once. A single user talking to an agent generates participant minutes for each participant in the room — the human and the agent both count — plus agent session minutes, plus egress if you are recording or streaming out. Agent minutes at around $0.01 dominate: at that rate the transport cost is a rounding error next to the agent meter, which is the opposite of what the headline participant rate suggests.
Sign up at cloud.livekit.io, create a project, and copy the API key and secret from the project settings. You get a project URL alongside them; the server SDKs use all three to mint access tokens. The Build tier is active immediately and self-hosting the open-source server is always an option if you would rather pay in ops than in minutes.
Not affiliated with LiveKit. Prices are reference estimates — always verify on the official pricing page.