LiveKit

New model just landed on LiveKit Inference: Google DeepMind’s Gemini-4 Argon! Drop it into your voice agent with a single line of code, run a test call, and hear the difference for yourself. Try it out now: https://lnkd.in/esQUDr_w

80% of Moss customers use LiveKit. There’s a reason. 👀 LiveKit gives you the real-time voice infrastructure. Moss gives your agent the context it needs to actually be useful. With Moss + LiveKit, retrieval happens automatically on every user turn, with results injected into the conversation before the LLM responds. No tool-calling step. No extra roundtrip. No dead air. And with sub-10ms retrieval, your voice agent can stay fast while still having access to the context it needs. See how it works: https://lnkd.in/encAP9mX

Join us with LiveKit next week at #LATechWeek to talk all things agents.

That's a wrap on #SFTechWeek for LiveKit! We kicked off the week with a Robotics Happy Hour alongside Dimensional. Our Head of Robotics 🚁 David Chen (and his co-presenter, Reachy Mini) spent the evening with builders, and researchers in physical AI discussing how real-time voice and video are bringing robots to life and how LiveKit powers physical AI in the real world. Then we closed things out with Voice as an Interface, co-hosted with Speechmatics and Aqua Voice. Dan Tacci built and shared a custom LiveKit voice demo, then joined Ricardo Herreros Symons (CRO, Speechmatics) and Finn Brown(CEO, Aqua Voice) on a panel about why voice is becoming the default way we interact with AI and why LiveKit stands out. The best part of the evening was the steady line of builders waiting to talk through their own voice agent use cases with our team. Huge thanks to our partners, our speakers, and everyone who came out. Next stop: #LATechWeek. We're teaming up with Modal for the AI Agents Speakeasy, an intimate, invite-only happy hour for the builders shaping the future of AI agents. See you in Santa Monica 🍸

Next week we're hosting an AI Agents Speakeasy with Modal during #LATechWeek: a cocktail evening built for the builders and founders shipping agents right now! Wednesday, Oct 14 · 6 to 9pm · Los Angeles RSVP below.

The worst way to learn your voice agent's limits is from a customer who just ran into one. LiveKit Simulations runs your agent against simulated users across every scenario, and hands you a pass rate before you ship. It scores conversations automatically alongside your pull requests; like a QA team for your agent. And when you're weighing a cheaper or faster model, you can test it against your own scenarios first and compare cost against pass rate before you switch to a new model. One platform, one bill, a lot fewer surprises after launch. Try LiveKit Simulations now, free through October: https://lnkd.in/gGRgtTTH

AssemblyAI Universal 3.6 Pro is now available in LiveKit Inference! What's new vs. 3.5 Pro: • 45% fewer wrong yes/no confirmations (52% fewer in heavy noise) • ~30% less background speech in transcripts • 14 new languages (32 total), with auto-detection and mid-sentence code-switching • Endpointing that holds the turn for phone numbers, codes, and emails, plus a new end_of_turn_confidence property • Drop-in upgrade: swap to universal-3-6-pro, same $0.45/hr Try it now > https://lnkd.in/ghHqKzgM

MAI-Transcribe-2-Streaming, MAI-Voice-2.1, and MAI-Voice-2.1-Flash from Microsoft AI are now available on LiveKit. MAI-Transcribe-2-Streaming debuts at #1 on the Artificial Analysis accuracy leaderboard. Combined with MAI-Voice-2.1-Flash, it gives developers what they need to build efficient and expressive voice agents. Try them out now: https://lnkd.in/g-mEi3qC

If you're building in robotics, you already know the good conversations don't happen during a keynote. Next week we're hosting a Robotics Happy Hour during #SFTechWeek with Dimensional, built for exactly that: short talks on where robotics and Physical AI are heading, then a full evening of food, drinks, and time with the people actually shipping this hardware. Tuesday, Oct 6 · 5:30pm · San Francisco Spots are limited! RSVP below.

Next week, we're getting into the hard parts of voice AI: latency, interruptions, turn-taking, and all the other stuff that never makes it into a demo video. Join us, Aqua Voice, and Speechmatics for a Voice as an Interface meetup during #SFTechWeek: talks and live demos from engineers at all three companies, then two hours of food, drinks, and conversation with the people solving this daily. Wednesday, Oct 7 · 5:30 to 8:30pm · San Francisco RSVP below.