Product

LiveKit Inference

Build voice agents with the leading AI models on the market. With LiveKit Inference, you can iterate quickly and swap models with a single line of code, with zero data retention by default.

Start building

Documentation

Illustration of isometric grid with LiveKit Inference elements

300,000+ developers

Billions of calls annually

300+ AI model integrations

Fast and reliable access to the best voice AI models

Deploy production-grade voice agents with the best-performing STT, LLM, and TTS models in the market.

Reduce end-to-end latency with global co-location of agents and models, dynamic routing, and provisioned LLM capacity.

A graph showing the end-to-end latency of an agent session.

Fast and reliable access to the best voice AI models

Deploy production-grade voice agents with the best-performing STT, LLM, and TTS models in the market.

Performance at scale

Reduce end-to-end latency with global co-location of agents and models, dynamic routing, and provisioned LLM capacity.

No Model Lock-In

Swap in the latest voice AI models from any of the leading providers with one account, one bill, and no new vendor contracts.

New

Gemma 4 31B on LiveKit Inference

1from livekit.agents import AgentSession, inference
2
3session = AgentSession(
4    stt=inference.STT(
5        model="deepgram/flux-general",
6        language="en"
7    ),
8    llm=inference.LLM(
9        model="google/gemma-4-31b-it"
10    ),
11    tts=inference.TTS(
12        model="cartesia/sonic-3",
13        voice="9626c31c-bec5-4cca-baa8-f8ba9e84c8bc",
14    ),
15)

Zero data retention

Your prompts, audio, and completions are never logged or stored. LiveKit Inference and every model provider behind it operate under a strict no-retention policy, so sensitive conversations stay private by default.

An image of a shield with a check on a grid.

Concurrency simplified

Manage your concurrency limits for all your voice AI models in one place.

Screenshot of the LiveKit Cloud inference usage UI.

Detailed observability

View turn-by-turn latency statistics and traces for inference requests in LiveKit Cloud to optimize agent performance.

The LiveKit Platform

Build, run, and observe
agents with LiveKit Cloud

Our end-to-end platform powers enterprise-grade voice AI
for customer support at global scale.

Explore the full platform

Resources to help you learn and ship

Read the docs

Gemma 4 31B: a faster, cheaper default for voice

product

Introducing LiveKit Inference announcement

Read the announcement blog

product

Prompt your agent to speak like a human

guide

FAQs

How do I access LiveKit Inference?

Any agent using LiveKit Cloud for transport has access to LiveKit Inference. Use the inference module in the Agents SDK to select which AI models your voice agent should use. To learn more, visit our docs.

What voice AI models are available on LiveKit Inference?

LiveKit Inference supports over 50 different voice AI models, including the latest models from OpenAI, Gemini, Cartesia, Deepgram, and ElevenLabs. For the complete list, see our pricing page. Looking for a strong default? See why we recommend Gemma 4 31B.

What are the concurrency limits on LiveKit Inference?

Each LiveKit Cloud plan includes a monthly quota of LiveKit Inference credits (billed based on model prices) and a limit on the number of concurrent requests. To learn more, visit our pricing page and our docs.

Does LiveKit Inference store or retain my data?

No. LiveKit Inference operates under zero data retention by default. Your prompts, audio, and model outputs pass through to generate a response and are never logged or stored, and the same commitment is contractually enforced with every model provider behind the service. Sensitive conversations stay private without any configuration on your part. For details on how we handle and isolate data, see our Security overview and our docs.

How much does it cost to access AI models with LiveKit Inference?

LiveKit Inference credits are billed by model type: LLM by tokens, STT by duration, and TTS by characters. Discounted rates are available for most STT and TTS models on the Scale plan. Preferred rates for high-volume usage are available with an Enterprise contract. To learn more, visit our pricing page.

How do I decide between using LiveKit Inference and LiveKit Agents plugins?

For fast and reliable access to the most popular AI models, use LiveKit Inference. If you want to bring your own model or use a different provider, use the LiveKit Agents plugins. To learn more, visit our docs.

Is LiveKit Inference only available for agents deployed to LiveKit Cloud?

LiveKit Inference is available to any agent using LiveKit Cloud for transport, including self-hosted agents and agents deployed to LiveKit Cloud.

Ready to build?

Start building a voice AI agent with a free account. Reach out to us if you're interested in custom pricing.

Start building

Contact sales

No credit card required • 1,000 free agent session minutes monthly

LiveKit Inference