One thing we've been excited about since the beginning of Gabber is Avatars Here we have Visemes animating the mouth on a ThreeJS rendered realtime AI powered VRM avatar

Can I build a realtime basketball coach that counts my shots and gives me voice feedback on my jump shot? Using Qwen3-VL 30B and a conversational AI, I found out - build tutorial at the end Link to try it below

Stumbled upon this video on youtube Word is getting out piped.video/watch?v=cXAkrPRm…

Introducing Gabber Realtime, multimodal AI inference + orchestration for Vision, Speech, and Language Models Build AI that can see, hear, talk, call tools, and do anything for $1/hr

Building real-time personal co-pilot that can see your screen and talk with you at @GabberDev.

A real-time learning assistant that sees your screen, talks, and learns alongside you. All open-source models (Qwen3-VL, Parakeet, Orpheus) wired together Homework, research, studying - all made easier Watch the video to see just how well it did

You can Video chat with Qwen3 Omni right now We're hosting the 30B Instruct and it's fast and pretty versatile

Want to build a simple realtime AI app in 5 min? Here's a tutorial on how you can create a conversational AI that runs on your laptop

Turned @Alibaba_Qwen 3-Omni into a screen-aware code copilot. It watches @dwyer_neil 's screen, catches issues problems, and suggests fixes. Ran on H100 (FP8 Dynamic Quant) via @GabberDev

@GabberDev is closing in on its first 1k GitHub Stars github.com/gabber-dev/gabber

New open-source multimodal models are wild. They don’t just read and write, but can see, hear, and think. Here’s @Alibaba_Qwen Qwen3-Omni (thinking mode) running locally on an H100. Smarter than 2.5, handles voice, tool calls, and real-time context. @GabberDev will set these models free.

Check us out @GabberDev we’re going to be hosting it by EOD

RT @jackndwyer: I think it's stupid that I have to write things when AI can now see and hear So I built an AI that: - talks to me - contin…

Why is it that all conversational voice AI agents are one-on-one? And what if I told you that they don't have to be? 2+ people. 2+ AI - that's now possible Current voice AI platforms are LLM sandwiches. You take your meat (your LLM), you wrap it in speech models (STT and TTS), you stack it together (with something like LiveKit). That's not to say being an LLM wrapper is bad, but more that you literally can't add participants without building special infrastructure to support it. i.e. Slapping a TTS and STT system around an LLM does not leave much room for innovation. You need new infra and new orchestration. And infra cannot be vibe coded. It requires real investment and is more akin to configuration than logic. We built a system that lets you build multi-participant conversational Voice AI and it was trivial using Gabber's new Multi-participant STT node and templating system.

Like Siri, but for your life. Tom Cruise approved ✈️

Amazon drivers keep peeing on my house I built an AI to ask them to stop

Can AI be your personal stylist? Using Qwen2.5-VL 7b, we found out

Experimenting with a pushup counter on @GabberDev to stress-test our state machine. Not perfect yet but getting there.