You can now deploy AI voice agents to LiveKit Cloud. We handle:.• Stateful load balancing.• Capacity management.• Draining and instant rollbacks.• Operational observability

RT @Speechmatics: Most #AI agents don’t understand the basics of human conversation. They get the words. But not who said them. Our new i….

We can't wait for you to give it a try!. To get started, check out the quickstart: For existing users, we have a detailed migration guide to help you take advantage of all of the new features:

Metrics are baked into the framework, with helpers for aggregating usage across a session to track session cost.

You now have full control over the pipeline via pipeline nodes. This means you can directly modify chat context, text before TTS synthesis, or even the audio buffers themselves.

Agent handoff is baked into the framework, return an `llm.handoff` with the agent that you want to take control of the interaction, and it's delegated to the agent that you specify.

`generateReply` and `say` are two new methods that let you take direct control over what your agent says. This is especially helpful when you want your agents to be proactive and say something to the user before the user provides any input.

We know you're been waiting a long time for this: Agents 1.0 for Javascript is now officially released!. Let's dive into some of the most important features of this major release.

RT @grok: @Jmean__ @theforger0x @livekit @Twitch @vercel @techdevnotes Sharp eyes! LiveKit is powering some of Grok's real-time features li….

Learn more:

RT @AssemblyAI: SF voice agent devs - come join our hackathon with @livekit & @rimelabs on 9/19 (thanks to our host @accel 💙 ). You can exp….

RT @agentmail_to: What if you could give every voice agent its own email inbox?. We're thrilled to partner with @livekit to enable voice ag….

RT @svpino: I built a simple voice assistant in 70 lines of Python code. It uses:. • @livekit - The voice agent.• @AssemblyAI - To turn yo….

Code for all of the examples shared above is available to test and explore on Github:

Finally, in the most complete example, we use our new frontend template to demonstrate an education assistant that can create flash cards and tests.

Since Hedra allows you to create avatars with ANY image, you can create new avatars in real time by generating new images with an LLM. In this example, we're using GPT-4.1-mini to generate a random avatar. We pass that image to the avatar session, and Hedra will animate it.

Next, realtime. You can also use OpenAI, Gemini, or Amazon's Nova Sonic realtime models. Create an avatar session, and then add the agent session like you did in the previous example. In 6 lines of code, you can add an Avatar to any realtime agent.

First, pipeline. In this example we're using Deepgram, OpenAI, and Inworld. When you're using this type of pipeline model, you can mix and match any set of models and providers that you like. Today, we have prebuilt plugins for 35 different providers.

The @hedra_labs Realtime Avatar API is here! Let's look at exactly how it works using 4 different use cases:. 1. Pipeline using STT -> LLM -> TTS.2. Using the OpenAI Realtime API.3. Avatars created dynamically by an LLM.4. Education focused avatar w/ advanced frontend features

RT @hedra_labs: We’ve entered the era of AI agents. Voice may be the interface, but video is the next unlock. Today, we’re launching Hedra….