Never let a signup go cold again.
Turn emails into revenue with Brew. No credit card, free credits to try.
Turn emails into revenue with Brew. No credit card, free credits to try.
livekit.io · newsletter
Explore this email design and adapt it to your own brand. Review the copy, links, and offer before sending.
dev-newsletter-v2-rounded-nocaption@2x
Hey LiveKit Devs,
Lots to cover in this roundup. Let's get into it.
newsletter separator
Product Changelog
Introducing Gemma, our latency-optimized LLM tuned for voice
The smartest models are usually too slow to feel natural in a live conversation. Gemma 4 31B is our answer to that: an LLM deployment tuned for real voice deployments: long system prompts, lots of tools, & almost no latency budget. It provides the same answer quality at 5.2x lower latency and 6x lower cost, and is now the recommended default LLM. Check out this blog post for benchmarks.
gemma-4 social img1 (1)
Make your agent sound human with expressive mode
Flat, over-enunciated delivery is one of the fastest ways to break the illusion that someone is actually talking to you. Expressive mode gives you control over how lines land, with defaults that sound good before you even tune. It works across Fish, Inworld, Cartesia, and xAI, and both the Python and TypeScript starter apps support it out of the box. For more info, check out our docs, Github example, blog post and video.
expressive mode img
You now pay by the second
Agent session minutes and recording, WebRTC connections, and SIP connections are all billed per second now, with a 10-second minimum increment. It applies to every plan and every account, including enterprise contracts, and it's automatic and already on your bill for August. See the blog.
Zero data retention by default on LiveKit Inference
Every model you call through LiveKit Inference is configured not to retain or train on your data, across STT, LLM, and TTS. It applies to every plan, with no flag to set and nothing to configure in your code. Russ published a Linkedin article on where we stand with privacy and data, and a technical breakdown of how ZDR works alongside it. Full provider list in the docs.
Redact PII before it reaches your observability data
Voice agents hear things form fields never collect: card numbers, addresses, dates of birth, account credentials. PII redaction for Agent Observability strips that out before anything is stored. It covers 41 PII types across ten groups. Read the blog.
Unity SDK 2.0
Our Unity SDK is out of developer preview and ready for production. Bring realtime audio, video, and data into your Unity application and connect it to LiveKit Cloud or your own server. Read the blog or watch Max's video on YouTube to learn more.
unity newsletter img
Video frame metadata
You can now attach metadata to individual video frames and have it travel with them through the pipeline, so per-frame context stays aligned with the image it belongs to instead of arriving on a separate channel and needing to be matched up after the fact. Read the blog post.
Async tools: keep talking while work runs
When a tool call takes real time, the default behavior is dead air, and dead air is where users start talking over your agent or hang up. Async tools keep the conversation going while work finishes in the background, then folds the result back in when it's ready. Read the blog or watch Jesse's demo for more details.
Content & Community Highlights
Build a voice agent that won't go off script
An agent that does its job well will just as happily do things you never scoped it for, whether that's writing a caller a poem or getting talked into a refund that was never owed. Darryn's guide covers the layers that keep it in bounds: prompt structure, rules enforced in code, an independent moderation check, and evals to catch what slips through.
See every avatar before you pick one
Choosing an avatar provider used to mean signing up for each one to find out how it actually looked in motion. Our docs team recorded demos across all 15 supported providers, so you can compare them side by side and then follow the setup guide for whichever one you want. Start with the avatar overview.
Build a RAG agent with MongoDB Atlas Vector Search
Most voice agents can’t remember you. Our new tutorial shows how to give your voice agent access to a private knowledge base using MongoDB Atlas Vector Search. Use vector search as a tool the LLM can call mid-conversation to retrieve relevant context and ground its answers. Watch Jesse's tutorial for a step-by-step walkthrough, or dive deeper into our docs.
Where to apply noise cancellation in your voice pipeline
Real microphones pick up fans, music, and the person talking three feet away, and all of it degrades transcription and turn detection. Our latest guide covers how noise cancellation works in LiveKit, where to apply it, and how to choose between background noise suppression and voice isolation across the Krisp and ai-coustics models.
LiveKit in the Wild
Run a voice agent entirely on Grok models
We built a patient intake agent with SpaceXAI that runs end to end on Grok voice models. Grok STT → Grok 4.3 → Grok TTS, cascaded through LiveKit Inference. Three model strings in one AgentSession. No separate API key, no separate billing, ZDR on every hop. Read the announcement tweet or give it a try for yourself.
Speechmatics' latest STT model available on LiveKit Inference
Linden, Speechmatics' new speech-to-text model built for voice agents, is now available through LiveKit Inference, no separate API key, account, or invoice required. Check out the release blog or start building with Linden today.
Gemini 3.5 Transcribe live on LiveKit Inference
Gemini 3.5 Transcribe Live is now available in LiveKit Agents, bringing LLM-based realtime transcription built for alphanumerics, domain-specific vocabulary, multilingual speech, and code-switching. Give it a try today on LiveKit Inference.
SF & LA Tech Week Events
📅 October 2026
📍 SF & LA
We're hosting three events during SF & LA Tech Week this October. On Oct 6, we're teaming up with Dimensional for a Robotics Happy Hour in San Francisco. The next night, Oct 7, we're collaborating with Aqua Voice and Speechmatics for a Voice AI Meetup. Finally, we'll head south for our AI Agents Speakeasy with Modal on Oct 14 in LA. Sign up before they sell out!
AI Engineer New York
📅 October 12-14, 2026
📍Sheraton New York Times Square Hotel
In the New York area and want to connect with us? We will be exhibiting at the upcoming AI Engineer New York event where we'll have a booth for hands-on demos, a dedicated speaking session on the agenda and more. We hope to see you there!
newsletter separator
You are receiving this email because you opted-in to receive email from the list LiveKit Developers from LiveKit
LiveKit, 369 Sutter Street, San Francisco, CA 94108, USA
Unsubscribe
Manage preferences
Let's build together
Join our developer community
Follow us on X
Contribute on GitHub
Subscribe to YouTube
LK_icon_darkbg
