What problem is LiveKit and Daily.co solving?

Building real-time voice and video applications presents unique challenges that traditional HTTP-based architectures cannot adequately address.

These applications require:

• Low-latency bi-directional communication
• Reliable connection handling
• Complex WebRTC implementation
• Media stream management
• Network traversal (STUN/TURN)

Normally, developers have to write thousands of lines of boilerplate code to handle these requirements.

LiveKit and Daily.co are open-source libraries that abstract away some of this complexity, making it easier for developers to build real-time applications like voice and video agents.

Takeaways

• If you have millions of concurrent users: use LiveKit.
• If you want a completely open-sourced solution: use LiveKit.
• If you need the absolute fastest latency: use LiveKit.
• If you are more focused on LLM pipelines and want an opinionated solution: use Daily.

What are the main features of LiveKit and Daily?

Both LiveKit and Daily are built on WebRTC and provide a horizontally-scaling WebRTC SFU (Selective Forwarding Unit) that allows applications to support large numbers of concurrent users.

They also both simplify the implementation of WebRTC by offering consistent API primitives across platforms.

Finally, they both offer a number of advanced features that are not available in other WebRTC solutions.

LiveKit overview

LiveKit: Flexible Deployment Options

You can either self-host LiveKit or use their managed cloud service.

1. Self-Hosted Option: For organizations requiring complete control over their infrastructure, LiveKit can be self-hosted.

2. Managed Cloud Service: For teams preferring a hands-off approach, LiveKit Cloud provides a fully managed solution.

LiveKit Agents

LiveKit recently released LiveKit Agents, a server side framework that simplifies building voice and video agents with realtime capabilities.

One of the nice things about LiveKit Agents is that they already have “sandbox” frontends that you can instantiate. These sandboxes automatically connect to your agent that is running either locally or is deployed somewhere like Render or Modal.

See our guide on how to deploy LiveKit agents on Modal.

LiveKit integrations

LiveKit’s ecosystem is particularly notable for its extensive integration capabilities. The platform offers plugins for various AI and media processing services, including:

• Cartesia for spatial audio
• ElevenLabs for voice synthesis
• OpenAI for real-time transcription
• DeepSeek for LLM

Daily.co overview

Daily is the company behind Pipecat, an open-source library for building real-time voice and video agents.

A core primitive of Pipecat is the Pipeline, which is a sequence of processing stages. Each stage can be a video processor, audio processor, or data processor.

For each processing stage (e.g. transcribing speech, getting a response from the LLM, generating audio), Pipecat takes a streaming approach, i.e. it processes data as it becomes available, rather than waiting for all the data to arrive.

Pipecat deployment options

Similar to LiveKit, Pipecat offers both a self-hosted and a managed cloud service.

1. Self-Hosted Option: For organizations requiring complete control over their infrastructure, Pipecat can be self-hosted and used to define your own pipelines.

2. Managed Cloud Service (Daily Bots): Daily Bots is “hosted” Pipecat.

Product experience comparison

A lot of the differences between the two libraries stem from the fact that LiveKit started out with a focus on real-time video and audio, and has since added on the agents/AI component.

Daily, on the other hand, started out with a focus on LLM pipelines, and has since added on the real-time video and audio component.

They are both quite battle-tested: LiveKit powers OpenAI’s real-time API while Daily has worked with companies like Accenture and Nvidia.

In both cases, your agent is essentially making calls to a WebRTC server, hosted by either LiveKit or Daily.

LiveKit is in general more flexible, generic, and open-source, while Daily is more opinionated in its pre-built characters and voices, although it is also extensively configurable.

Pricing comparison

Pricing for LiveKit Agents is subscription-based. There are several tiers that come with different amounts of included minutes, bandwidth, and other features.

Pricing for Daily Bots is usage-based, at $0.02/participant minute. You also have to pay for your usage of the integrated services like ElevenLabs, Deepgram, and Cartesia, for which you are billed through Daily.

Performance comparison

Concurrency

LiveKit: Depends on your plan: Build plan is 100 concurrent participants, Scale plan is unlimited.

Daily: With a card on file, Daily Bots will give you up to 100,000 concurrent calls or sessions.

Latency

For real-time applications, the goal in terms of end-to-end latency is generally something like 500 ms or less. This latency is made up of a number of legs, for example:

• user’s voice from device to cloud (via WebRTC)
• transcription
• LLM inference
• speech generation
• output voice from cloud to user’s device (via WebRTC)

LiveKit claims to be able to be able to do the first and last leg (the WebRTC legs) in sub-100ms latency.

Daily doesn’t cite the WebRTC latency but with careful optimization (e.g. colocating LLM, TTS, and STT models on their own infrastructure) quotes end-to-end (voice-to-voice) latency in the 500 ms range.