LiveKit, a startup that provides software underpinning voice, video, and physical AI models, raised $100 million in a funding round that values the company at $1 billion.
The investment was led by Index Ventures and included Salesforce Ventures, as well as prior investors Altimeter Capital Management, Hanabi Capital, and Redpoint Ventures.
Building a full development lifecycle
LiveKit is developing tools to support every stage of building voice agents. On the development side, they provide client SDKs for various platforms and the LiveKit Agents framework. This framework helps developers manage tasks such as integration and conversation flow, including turn-taking.
They also offer Agent Builder, which lets teams start with pre-made templates instead of writing code from scratch.
Testing voice agents can be tricky because AI behaviour can be unpredictable, so simple tests aren’t enough. LiveKit emphasises the need for developers to assess agents statistically by analysing many conversations.
Their platform supports unit testing, OpenTelemetry tracing, and large-scale simulations, and they collaborate with partners like Bluejay, Hamming, and Roark.
LiveKit addresses the challenges of live conversations, as they can be unpredictable in terms of duration and demand. To manage this, they use serverless agents and have established a global network of data centres designed for low-latency voice and video calls, handling billions of calls each year across various platforms.
To improve phone-based voice agents, the company is partnering with carriers to connect directly to the public telephone network, as performance relies on effectively coordinating multiple models, such as speech-to-text and text-to-speech. Issues like distance or system delays can negatively impact the user experience.
To tackle these challenges, LiveKit has created LiveKit Inference, which directs traffic across multiple model providers while monitoring in real time. They’ve also begun hosting models in their own data centres to bring inference closer to the agents for better performance.
For teams running voice agents in production, LiveKit has added observability tools designed specifically for live conversations. These tools help answer practical questions such as how quickly calls are answered, what the agent heard, whether users asked for a human, how latency changed during a session, and whether the right tools were triggered.
Unlike traditional web apps, LiveKit argues that voice applications need an entirely different stack. Conversations are real-time, stateful, and can last minutes or hours. Agents have to listen, think, and respond continuously, while keeping context throughout the session. That changes how software is built, tested, deployed, and monitored.
What’s next?
With the new funding, LiveKit aims to make the process of building and scaling voice AI as simple as developing and scaling web applications. This initiative is designed to help them accelerate progress as voice technology becomes an integral aspect of how people interact with software.
