Ey guys, just submited a PR to add ref_audio support to Voxtral TTS in Livekit Agents https://github.com/livekit/agents/pull/5278

Hi! I built an Expo plugin for LiveKit screen sharing. It handles iOS Broadcast Extension setup automatically. No more manual Xcode configuration, just add to your plugins:

"plugins": ["expo-livekit-screen-share"]

GitHub: https://github.com/arekkubaczkowski/expo-livekit-screen-share npm: https://www.npmjs.com/package/expo-livekit-screen-share

Hi folks! We launched traceAI on Product Hunt today! it's open-source LLM tracing repo that understand GenAI semantics and gives you real visibility into LLM calls, prompts, tool use, etc. If you have a sec, an upvote would really help us out! There's an exclusive livekit support which can be leveraged using

traceai_livekit

for livekit telemetry in any platform of your choice https://www.producthunt.com/products/future-agi/launches/traceai

Among the more common problems my users complain about is what I call the "Be Quiet" problem where agents don't know how to mute themselves temporarily. It happens when the user tells the agent to be quiet for a minute while they have a short conversation with someone else. Even though the agent says it will be quiet, it doesn't stay quiet. The user could hit their mute button, which works if the agent is the only other participant in the room, but if there are other humans or agents, this isn't an option. I built a solution that I'm testing and you can see back-to-back demos first without the "Be Quiet" feature and then with it: . You can read about the architecture here: https://eniveld.substack.com/p/your-voice-ai-agent-doesnt-know-how

Hey - Prashant here, building Rapida - open-source voice AI orchestration in Go. Similar space but different approach - Rapida has a native SIP stack, no SIP-to-WebRTC bridge needed. More about full ownership and control of the voice pipeline. Provider-agnostic, swap LLM/STT/TTS without code changes. Repo: https://github.com/rapidaai/voice-ai Docs: https://doc.rapida.ai/introduction/overview If anyone takes it for a spin, would love feedback. What breaks, what’s confusing, what’s missing.

Hi everyone! I’m excited to share what I’ve been building over the past 9 months, a real-time AI avatar platform (B2B API SaaS). The latest version is in a great place and I’d really appreciate any feedback from this community. You can try it out here: 👉 https://agenthuman.com Also, I’ve opened a PR on the LiveKit agents repo and would love your thoughts there as well: 👉 https://github.com/livekit/agents/pull/5340 Thanks in advance! 🙏

Salut, Quelques nouvelles de n8n-as-code : J'ai eu le plaisir d'être invité par n8n dans un épisode de leur podcast :

Hey everyone! Shiv here, building Quraite. After building voice agents ourselves, we learned that standing one up is easy; getting it to work reliably is the hard part. So we built Quraite to help simulate users, evaluate voice agents, and iterate faster. We support phone numbers, Vapi, and ElevenLabs, and LiveKit integration is on the way. We'd love to connect with voice agent builders to understand your development lifecycle and explore a design partnership. Feel free to reach out! https://quraite.ai/

Hey everyone! 👋 Voiceblox 0.7.0 is out. Two highlights worth mentioning here: - 🎙️ Voice picker with live preview — Browse and preview TTS voices directly in the builder before deploying. Click play, hear it, pick the right one. - ☁️ LiveKit Inference support — You can now choose LiveKit Inference as the compute backend when deploying. No need to run your own Python worker. Also in this release: - SIP Transfer can now disconnect right after the transfer completes. - Interruption mode simplified to a plain on/off toggle. Check it out: https://voiceblox.ai GitHub: https://github.com/voiceblox-ai/voiceblox

Hello everyone! We've been building Betula on LiveKit for a while now — wanted to share what we just shipped. Meet Lucy — our AI voice assistant for professionals and small business owners. We let her announce her own launch, voiced entirely through ElevenLabs (v3 with audio tags for emotion and pacing). What she does: • Handles inbound/outbound calls (LiveKit + Twilio)- • Responds to texts via SMS bridge (we have to still use Twilio since our agents need to receive texts) • Sets reminders, does daily briefings (stocks, news, weather) • Delegates tasks — finds local businesses, gets quotes • Persistent memory across all interactions She comes with her own phone number. Users can call, text, or list her as their assistant. Stack: LiveKit agents, Twilio, ElevenLabs, Deepgram, OpenAI, Cloudflare 🎥 Watch Lucy introduce herself →

Just published a book that builds LLM architectures from scratch in PyTorch — from Transformers through GPT-2 and Llama 3 to DeepSeek's MLA and MoE. Covers KV cache, GQA, and inference optimisation in depth. Sharing here since a lot of real-time AI work builds on these foundations. All code is open source. GitHub: https://github.com/S1LV3RJ1NX/mal-code Book: https://leanpub.com/adventures-with-llms

Built an agent that manages a tmux session so it can be used to add voice to control to Claude from anywhere. Also added a mobile app and an http tunnel so you can test your web apps while remote. https://www.linkedin.com/posts/yepher_control-claude-with-your-voice-from-anywhere-activity-7451298526979899392-i0A1/

Hey Beautiful People, I built OpenThread (https://openthread.me) - a community for sharing, discovering threads from Claude Code, Codex and other AI tools. Think of it as a place to find the prompts and workflows that actually work, instead of screenshots scattered across Twitter. The thing I'm most excited about is the Claude Code plugin. Install it and you can share any conversation with one command: https://www.npmjs.com/package/@openthread/claude-code-plugin npm i -g @openthread/claude-code-plugin /ot:share A few things I'd love feedback on: The privacy masking — what else should it strip by default? Community structure — right now it's topic-based (e.g. "Coding with AI"), but I'm considering tool-based or workflow-based splits. Happy to answer anything about the architecture or the rationale. Built with couple of folks, still rough in places, and genuinely want it to be useful rather than just another feed. Please show some love. Thanks. Parvinder

Hey everyone! I’m Visharad from thesys.dev. We are doing some really fun stuff with generative UI. Would love to connect and collaborate with folks here and contribute to the community! https://www.openui.com/

This is really good am working on the same thing but for Android

We have a webinar tomorrow for Voice Agent Webinar with Simplismart and LiveKit Come join us https://luma.com/jg93tpel

Hey folks 👋 We just open-sourced Future AGI - infrastructure for AI agents that improve themselves autonomously. The shift: Most teams stop at observability and evals. We built the full loop: agents that get better from their own production data, automatically, continuously. What we built: • Adversarial simulation (agents stress-test themselves before hitting prod) • Real-time self-correction (learned classifiers block bad outputs mid-conversation, sub-50ms) • Autonomous improvement pipeline (every failure becomes training signal automatically) • Full transparency (inspect exactly how your agent is learning and evolving) Ran this in production for months. Today: fully open source. Why open source? Can't ask people to trust a closed system that autonomously improves their AI. You need to see the learning mechanisms. Also - self-improving AI is foundational infrastructure. Every prod team will need it. Better built once, in the open. https://github.com/future-agi/future-agi Self-hostable. Full stack. For teams ready to build agents that actually learn. Drop questions below - happy to go deep on any part 🧵

Hey @most-diamond-37100 do you support WHIP streamimg?

Hey everyone! I built Quest Learning, a smart AI app that acts as your personal tutor for anything you want to learn. I designed it so you can just type in any topic or upload your own textbook, and Quest instantly breaks it down into a custom, 100-step course. But I didn't want it to be just another reading app, I built it so you actually talk to it. I added an AI Voice Tutor that speaks with you and asks questions to make sure you truly understand the lesson. At the same time, I built a Smart Board that draws diagrams and notes on your screen in real-time, so you can see exactly what the tutor is explaining. The Tech Stack I Used: Backend: I Used Spring Boot (Java) + PostgreSQL with PGVector and Spring AI to handle RAG (Retrieval-Augmented Generation) for massive textbooks. Frontend: Android & iOS (built with Native Kotlin & Swift) utilizing LiveKit's mobile SDKs for real-time WebRTC streaming and custom UI rendering for the interactive whiteboard. Voice & Visual Sync: I'm using LiveKit to handle the low-latency voice dialogue (Gemini Live). Simultaneously, I use LiveKit Data Channels to push JSON packets to the frontend, drawing on the Smart Board in perfect sync with the AI's audio. My Ask: I’m opening a Closed Beta and I'm looking for two types of early testers: Technical Testers: To help me stress-test the app. I’d love your honest feedback on the voice interruption latency (VAD) and how smooth the audio-to-visual sync feels on your device. Students & Avid Learners: To test the actual teaching! Upload a textbook or a topic you're currently studying, and let me know if the Socratic Tutor actually helps you understand it better. Join the Beta here: Beta Link (Note: Once you've filled out the form, please DM me or drop a comment below so I can personally add you to our private Discord server!) (Check out the quick demo video below!)

https://loqal.digital Loqal, built by an unusual alliance of developers from around the world, lets speakers of 177 languages learn in their own language, making for over 30000 language pairs. LinkedIn: Aviah Morag

Hey everyone 👋 I’m Aditya — I’ve been building in the LiveKit ecosystem for over a year, and I just shipped the beta of Meridial (meridial.dev), an embedded support suite for SaaS products. It puts a support agent directly inside your product to answer user questions in real time. You can jump into conversations anytime, talk to users live, and even take control of their screen when they’re stuck. I built this after repeatedly seeing the same issue with complex UIs (think AWS/GCP-style tools) — users default to ugly docs or email because they can’t figure things out. Meridial brings support (you + agents) into the product itself, where your users are. I’m looking for a few pilot customers to test this in production. You’re a good fit if: • Your product has a complex UI • Users frequently reach out for help via Slack or email • You want to stay close to users without scaling support headcount In exchange for feedback, you get one month free. If this sounds relevant, DM me — happy to walk you through it. https://meridial.dev

Wrote up how I built pause/resume for LiveKit agents: https://eniveld.substack.com/p/pause-is-a-listener-concern-building The TL;DR is that this is what I think Issue #3740 actually wants — implemented as a Web Audio worklet ring buffer on the client rather than session-level state on the agent. Multi-participant safe by construction (the agent never pauses; only the listener stops listening). 52-second demo video and a live demo link at the end of the post. Feedback welcome.

Hey everyone! 👋 Voiceblox 0.8.1 is out, and the big news is that it's now hosted at voiceblox.ai. Sign up, build, and deploy your LiveKit voice agent without cloning a repo or running anything locally. Two new providers worth flagging here: - 🎙️ Gemini STT — Now available in the builder. - ⚡ xAI STT and TTS — Plus updated LLM and Realtime models. Check it out: https://voiceblox.ai

Hi this is exciting. @fresh-flag-73780 But want to know if this also works for Indian regional language aswell?

Can a technical article about LiveKit be fun to read? How about an article that begins by discussing how to use LiveKit Data Tracks to let crows (yes, the bird!) join your LiveKit call? Then I generalize the concept to other domains like remote medicine. https://eniveld.substack.com/p/when-crows-join-your-livekit-call

You all may have seen me in the #C088ZNU7QQ5 or #C07FY8WHGPM communities. We’ve been building on Livekit for a while, and today we launched our product for the first time. We are so thankful to the Livekit community, and proud to say the foundations of our product are powered by Livekit. We’ve just launched, please give it a like and repost! https://www.linkedin.com/posts/ruchirbaronia_introducing-frontdesk-the-worlds-first-ugcPost-7460402012308856832-4d5b/?utm_source=share&utm_medium=member_desktop&rcm=ACoAACtD9agBM08C8ViKJDFLEEc-roHwMlMRIW4 https://x.com/BaroniaRuchir/status/2054645354327408687?s=20

Hey 👋 — I built a voice/video calling app with the LiveKit RN SDK, and I extracted the native call UI part into a standalone Expo module: expo-callkit-telecom. It wraps CallKit on iOS and Jetpack androidx.core:core-telecom on Android with one typed API, plus a config plugin and native VoIP push parsing (APNs + FCM). CallKit+PushKit integration was one of the harder parts of building a VoIP app so I hope this proves helpful to others. Docs (with iOS + Android demos on the home page): https://mfairley.github.io/expo-callkit-telecom/ Repo: https://github.com/mfairley/expo-callkit-telecom

Cool to see our robotics team cooking: https://x.com/southpkcommons/status/2057876338753900946?s=46&t=zHy_g8wAVw-P8FzLY65yQQ