Object selection using local AI

🚁 HyperDrone: Drone simulator at >10M FPS ⚡️ HyperLux: Massively parallel ray tracing for RL 🔌 Any GPU: Nvidia, Apple Metal, Vulkan, WebGPU 🦾 Sim2Real transfer of an RGB policy ✅ Accepted at CoRL 2026 Check it out: pip install hyperdrone Links 👇

distilled a model specifically to play mario, a 9M model on the M5 Pro ANE (0.5 ms per move). ~1000× smaller. Here it is playing 23 levels in a row 👇 Jev on Core ML: huggingface.co/FluidInferenc… 9M Mario Core ML model: huggingface.co/FluidInferenc… Swift runtime + demo: github.com/FluidInference/Fl…

Prompt to 3D character animation, now with high fidelity hand + finger tracking✌️ FLUX 3 Max video → WiLoR for hands + GVHMR for body tracking → any animated character! open source: github.com/blendi-remade/pro…

I have three generative AI models running inside a web browser. On my phone. Not a web page calling AI APIs. The actual models are running on the page inside the browser. I replaced Gemini Live in my RTS game prototype with an entirely on-device AI Commander. One model listens to my voice. A second, fine-tuned 350M parameter model understands my orders. A third speaks back. The game's own systems handle the strategy and execution. I can tell it to build defenses, move units, ask about the battlefield, or change my mind mid-conversation. All without a single inference request going to a server. It works on my Mac. It works on my Pixel 11. The decisions take roughly half a second, with no measurable impact on the game's frame rate. But here's where this gets really interesting: in a game built around generative AI, where using the AI is native to the experience, I just eliminated the cost of running that AI for every player. Normally, the more people play, the more they interact with the AI, the more the inference costs. Not anymore. The models ship with the game. Every player brings their own compute. They can interact with the AI as much as they want, and the inference cost stays at zero. No AI subscriptions. No API keys. No usage limits. No inference bill for me or the player. And because it's all running in a browser, I can deliver that experience across desktops and phones, anywhere really that has enough compute to run the game, all without building separate native applications. This fundamentally changes the unit economics of publishing an AI-native game. But I think the implications go a lot further than games. What happens when generative AI stops being a service you rent, and becomes software you can simply ship?

CoWork OS 0.5.61 in 95 seconds, recorded in the real app with a local model (qwen3.6:35b on Ollama). Install in one command, bring your Hermes agent, approve every action, give bots their own computer, and see only what needs you.

The iPhone Duo is out today, and @detailapp's Auto Edit runs on it entirely on device. Detail edits a 10-minute podcast recording in 8.4 seconds: a full edit plus a dozen shorts, each with a title and its own transcript. Six of our models do the work: Ear detects the language, Voz transcribes, Clear cleans up the audio, Uhm cuts the filler words, Clips picks the best moments and Title names the clips. desertant.com/models

🚨 introducing hobbylm-1b computer use -- fully offline, open weights & we made it 13x faster: 3.4 s → 0.26 s per decision. my 1b model booked a call for sunday 7pm while wi‑fi was off. it found the name in contacts, worked out which sunday, set the time in my time zone, asked y/n, then added the event to calendar with no internet and no cloud, on a macbook air. what else it does live offline as per tested: "split 600 among 4 people" → 150 in calculator "make my screen bright" → brightness up "sound is harsh, turn it down a notch" → volume down "mute everything" → muted "turn on dark mode" → dark mode on "bring back the light theme" → light mode on "i need to write something down" → notes opens "need papers from documents" → documents folder opens "seven minute timer for eggs" → 7-minute timer set "grab a screenshot" → saved to the desktop "what's the weather tomorrow" → "i can't do that one offline." yeah, how a 1b model covers this much in offline: 14 mb adapter on our own pretrained base, with a bit of fine-tuning, run fully on-device. thanks @hariharanms95 for pulling this off weights, code and every test result: huggingface.co/Hobbylm/hobby…

Nemotron 3.5 Bolt E4B creating PowerPoint on My Local AI about Nemotron 3.5 Bolt E4B 😀 apps.apple.com/us/app/my-loc…

WholeBoard now sims the board inside its enclosure. ESP32-C3 board, 333 mW: 42.6 °C in open air, 52.0 °C sealed in a PLA box. Add lid and side slots and it drops to 50.4 °C. Still in spec in a 40 °C room. Sealed or vented for yours? github.com/DingoOz/WholeBoar…

you can run small AI models on cell phones .. and build games on top of them ..

Five months ago I started building a radio station for my own music library. Today it has an AI DJ that picks songs and talks between them, takes requests, runs fully local, and has iOS, Android and desktop apps. Strangers are running their own stations on it. Solo, open source, MIT. Made an intro video. Would love honest feedback from people who’ve shipped things: does it make sense in the first 10 seconds?

I think I’m going to release the projection texture software for free. I don’t really feel up to handling customer support for a paid version. The video below shows a test of the projection texturing system on de_dust2. The software can texture outdoor environments too, but so far I’ve been getting better results with interiors.

Decision 2.0 is now in FluidUse 🍎 Kai 0.6B + Eos 0.8B in native Swift, on-device on an M5 Pro. Demo: 1,000 GitHub issues labeled live, 5 decisions per issue in one call (type, owner team, priority, flags). ~15 issues/sec. 🛠 github.com/FluidInference/Fl… 🤗 Kai: huggingface.co/FluidInferenc… 🤗 Eos: huggingface.co/FluidInferenc…

🚀 Meet MAI-Gym: an open platform for GUI agent evaluation & RL across mobile, desktop and web. 20,000 parallel environments, automated app/task synthesis, and open code, data & weights. 🔥 Our 9B agent scores 78.4% on MobileWorld and 76.2% on OSWorld-Verified.