Ian Ballantyne

Introducing the newest Coral board, for efficient, on-device AI! Check out the demos in the video: - On-board speech translation - Natural language controlling hardware - Vision & sound generating music

@badlogicgames Cool cool, let us know what you find! Being able to test different configs quickly was part of my requirements. We use an OpenAI-compatible endpoint here pointing to llama-server, but a quick config change and I can point to any model / inference stack. Really helped iterate 🙌

@IanBallantyne aye, brain runs on laptop, sensors/UI on the phone. i got everthing but the llm working locally. still shopping around for local models. gemma is high up on my list!

the robot's brain is done. pi + elevenlabs plus phone sensors/cameras + memory system. will build a cardboard chassis for phone/electronics tonight, put it on top the robot legs, then the MVP is done love living in the future.

@badlogicgames Nice! Is this running on the laptop directly or is it streamed? The latency for your vision processing + memory is pretty good already. Our goal is to have the whole stack offline on the hardware, is that a requirement for you?

@IanBallantyne similar stack! silero VAD -> parakeet -> undisclosed LLM :p -> Kyutai TTS

@badlogicgames You can also interrupt the model, which makes it more interactive.

@badlogicgames It uses Silero VAD -> faster-whisper -> Gemma 4 E2B -> Kokoro TTS (running on an M4 Max). You can swap out parts of the pipeline, but these are the ones I found led to a good trade off in quality and speed.

And a bonus here is one of the demos! #aiedgegallery

That's a wrap on Google I/O from the Gemma team 🙌 We had some incredible demos, talks and conversations with passionate Gemma users! Thanks all for connecting and sharing your stories 🤝. We're going to keep cooking. See you again next year #gemma #googleio

If you missed the livestream here is the recorded talk: youtu.be/oUtiZbrehrw?si…

Thanks to @o_lacombe @osanseviero @xenovacom @GlennCameronjr Gus, Xavier, Alice, Billy, Tom, Jenn, Derek, Alex, Michael, Allie, Trish, Rachel, John and everyone who helped bring this all together for the #gemmaplayground and talks!

@lmstudio @ForProduction @ggml_org Let us know if you need any help! 😉

@ForProduction The amazing team at @ggml_org and the community are working to bring MTP to more models

MTP is available in LM Studio 0.4.14. Sound on.

@xenovacom @thorwebdev @GoogleDeepMind Thanks for making such a cool demo

Transformers.js and Reachy Mini were featured on stage at Google I/O 2026! 🤯 Huge thanks to @IanBallantyne and the @GoogleDeepMind team for the shout-out! 🤗

At Google I/O today? Stop by the Gemma Playground and chat with our Gemma 4-powered Open Duck robot! 🤖🦆 Not on site? No problem! You can 3D print your own at home.

5️⃣ Gemma - Yup, just Gemma. We already have an incredible model in Gemma 4 💎💎💎💎, but the rest of these products will massively accelerate the work I'm doing. Imagine more demos, fine-tunes, examples, collabs, variants, evals, datasets and more! This is the productivity unlock and I'm ready for it🔓

4️⃣ Gemini Omni - The possibilities are incredible here. I don't think we realize yet what this means, both in terms of how we learn, present, interact, create and explore different mediums. This isn't just another generative media model, this is THE model 🤔

This I/O isn't about Google changing the world, it's about Google helping you change the world. 🌎 Now that the dust has settled on the I/O keynotes. Here are the top 5 things I'm excited for 💙❤️💛💚 👇

Introducing Gemini Spark ✨ It’s your 24/7 personal AI agent that helps you navigate your digital life, taking action on your behalf, and under your direction. 🧠 It runs on Gemini 3.5 and is built on @Antigravity, so it can perform long-running tasks easily in the background. ⏱️ And because it runs on dedicated virtual machines on Google Cloud, you don’t even need to keep your laptop open. 🧰 Spark will integrate seamlessly with Google tools, and soon with third parties through MCP. #GoogleIO