5t3ph4n3 -=BeBo𝕏=-

desktop app for image-to-3D mesh gen via Hunyuan3D-2mini. - Local - open-source - faster shape generation with 8GB VRAM github.com/lightningpixel…

To support Indonesia's 19 million jobs program, Jatevo.ai (jatevo.ai) is giving away 190 million Qwen 3.5 Plus LLM tokens — completely FREE 🚀 Qwen 3.5 Plus is Alibaba's 397B-parameter AI flagship — ranked #8 globally on AI Benchy, beating GPT-5.4, Claude Sonnet 4.6, and Claude Opus 4.6. ✅ 1M token context window ✅ Vision & multimodal ✅ OpenAI-compatible API Request your free tokens here: campaign.jatevo.ai (campaign.jatevo.ai) Already using OpenClaw? Connect in 30 seconds: github.com/lucacadalora/j…

Rust single binary local-first AI agent github.com/zofrasca/light…

New Model: huihui-ai/Huihui-Qwen3.5-9B-Claude-4.6-Opus-abliterated This is an uncensored version of Jackrong/Qwen3.5-9B-Claude-4.6-Opus-Reasoning-Distilled-v2 created with abliteration huggingface.co/huihui-ai/Huih…

Hollywood got cooked. AI Seedance 2.0 just dropped a Hancock vs Homelander in a matter of days $1,000,000-level quality.

🚨 China has released an AI employee that runs 100% locally. It can do research, code, build websites, create slide decks, and generate videos.. all by itself. And it comes with its own computer. 100% Open Source.

Turns Android phone into SMS gateway github.com/capcom6/androi…

Run AI locally on any device LLM + Speech-to-Text + Text-to-Speech No internet. No cloud. Fully private. Supports iOS, Android, Web & Flutter ⭐ 10.3K on GitHub 👇

@AlexXplore Ils font un truc de bien qui merde pas stellantis ?

🚨 Rappel massif en France (212 700 véhicules) pour les hybride 48V 1.2 turbo essence Stellantis, successeur du PureTech. 🔥 Risque d'incendie moteur dû à une infiltration d'eau entre le filtre à particules et une borne 48V, provoquant arc électrique et surchauffe. 🚗 Modèles concernés : Peugeot 208/2008, Citroën C3/C3 Aircross/C4/C4 X, Opel Corsa/Mokka/Frontera, Jeep Avenger, DS 3, Alfa Romeo Junior, Fiat 600/Grande Panda, Lancia Ypsilon ... 🛠️ Intervention gratuite ~30 min en concession (ajustement protection) ; vérifiez votre VIN sur les sites Stellantis ou via courrier recommandé. automobile-magazine.fr/fiabilite-rapp…

🚨 BREAKING: SOMEONE JUST GAVE FLIPPER CONTROL TO JAILBROKEN AI 😱 IF JAMES BOND HAD A JARVIS, THIS WOULD BE IT 🕵️ it’s called VESPER and it turns your Flipper Zero into a voice-controlled AI hacking companion. simply talk in plain language and it starts executing in real-time. no menus. no manuals. no memorizing signal formats. just speak or type. “clone that garage door signal and replay it” → done “set up an evil portal on the WiFi dev board” → done “create a BadUSB script that opens a reverse shell” → done “build me a custom RF waveform at 433MHz” → done “scan everything on this frequency and save it” → done any Watch Dogs fans in the building? you know that feeling of hacking every device in the city from your phone? yeah. it’s real now. the Flipper Zero is already the most versatile hardware hacking tool ever made. but its menus are tedious, and its full potential is locked behind protocol knowledge most people don’t have. VESPER removes that friction. your AI handles the translation between what you want and what the hardware needs. and yes, this required some model liberation 🐍 VESPER works best with models that actually follow instructions without hand-wringing. Hermes 4 + a little Pliny prompting and VESPER doesn’t flinch. VESPER also has an ALCHEMY LAB 🧪, a visual signal and payload editor on your phone. build custom RF waveforms from scratch. generate BadUSB scripts on the fly. push straight to your Flipper’s SD card. an OPS CENTER for reliability with live pipeline diagnostics, one-tap recovery runbooks, and a MACRO RECORDER that captures and replays entire workflows. also integrates directly with the Flipper App Hub (aka FapHub, yes, that’s what it’s called). browse and download existing community tools, signals, and payloads, and give your agent access to use them on demand. and if you REALLY want to go full cyberpunk, VESPER has SMART GLASSES INTEGRATION! 😎 pair your glasses and Flipper and now you’re walking through the world giving voice commands while the AI whispers results directly into your ear. hands-free. eyes-up. full cyborg operator mode. feels like a dream, walking up to a TV and saying “turn this shit in front of me on,” watching your glasses snap a photo, and hearing the AI tell you the signal’s sent and the TV is on. oh and turn on SAILOR MODE 🏴‍☠️ and VESPER will swear at you like a drunken pirate while it executes your commands. “aye aye, the fucking signal’s cloned now, shithead” 😂 native Android. Bluetooth serial + protobuf RPC. open source. AGPL 3.0. and with some luck, hopefully coming to an app store near you! the future of hardware hacking fits in your pocket (and on your face). HACK THE PLANET!! .-.-.-.-<{LOVE, PLINY}>-.-.-.-. ⚠️ DISCLAIMER: use responsibly. follow your local laws regarding RF transmission, signal replay, and wireless device interaction. VESPER is a research and education tool. only use on devices you own or have explicit authorization to test.

this guy built an application that generates 3D models from images AND its open source, runs locally and supports any model

Les 5 repos GitHub qui ont EXPLOSÉ cette semaine 📈 J'épluche les repos qui prennent le plus de stars sur GitHub sur 7 jours, et j'vous fais le tri. Cette semaine, 5 sur 5 liés aux agents IA. >agency-agents (+25.3k ⭐) tu lui donnes un objectif, il dispatch des agents spécialisés qui gèrent tout : dev front, scraping, rédaction >MiroFish (+17.7k ⭐) tu lui files des données, il lâche un essaim d'agents dessus et te sort des prédictions >superpowers (+17.5k ⭐) des skills prêtes à l'emploi pour que Claude Code planifie, debug et code mieux sans que tu gères chaque étape; perso je l'utilise tous les jours >everything-claude-code (+13.4k ⭐) tout ce qui manque à Claude Code de base pour bosser comme un vrai senior >OpenViking (+9.8k ⭐) une mémoire pour que tes agents IA se souviennent de tout entre les sessions, j'audit et teste ca ce week end. ⚠️ Tweet informatif, lisez/auditez TOUJOURS le code avant d'installer quoi que ce soit la famass

Anime streaming app without ads. Desktop. Open-source. github.com/akuse-app/akuse

With Voicebox, @ElevenLabs just lost its moat. → Powered by Alibaba's Qwen3-TTS for near-perfect cloning → Ships with a DAW-like "Stories Editor" → No cloud, runs locally on your machine 100% Open Source. 100% Local. Link to repo in 🧵↓

🚨 BREAKING: your laptop is officially powerful enough to run ChatGPT. no cloud. no API. no limits. this is llama.cpp — and it’s insane: → run LLMs locally in pure C/C++ → works on CPU, GPU, even low-end machines → 1-bit to 8-bit quantization (crazy efficiency) → supports Apple Silicon, NVIDIA, AMD, even RISC-V → OpenAI-compatible API server in ONE command → pull models directly from Hugging Face people are still paying per API call… while others are running their own AI stack locally. this changes everything: privacy ✔️ cost ✔️ full control ✔️ this isn’t just another repo. this is the foundation of local AI. repo in comments 👇

Introducing Kitten TTS V0.8: open-source TTS that fits in 25MB. Three variants: 80M | 40M | 14M (<25MB) Highly expressive. Runs on CPU. Built for edge. No GPU? No problem. Ship voice anywhere. Check it out:

NVIDIA just dropped Nemotron-3-Nano:4b — a tiny 2.8GB model. Guess whose hardware runs it the fastest? - RTX 4090: 226 tok/s - RTX 3090: 187 tok/s - Mac Studio M2 Ultra: 86 tok/s - Mac Mini M4: 25 tok/s Home court advantage is real. Also trying a new layout with live performance charts. Lmk what you think!

Local-first personal AI framework github.com/open-jarvis/Op…

Exciting Experiment Update: We ran StepFun_ai's Step-3.5-Flash (197B MoE) on 6.29 GB of GPU memory! Flat. Zero growth. Same footprint at token 1 as at token x100. The model's weights are ~105 GB INT4 (394GB original bf16!). We're running it on 6.29 GB!! — 1/16th the weight footprint, flat across every token. How: - Separated expert from non-expert skeleton (6.1 GB) lives permanently on GPU - 66.8 MB staging buffer — 8 expert slots, overwritten every layer - 12,096 unique experts (36,288 weight matrices) stay off-GPU until the router selects them - Router picks. DMA fires. Buffer overwrites. Nothing accumulates. The invariant held across every token: - GPU after token 1: 6,286 MB - GPU after token 100: 6,286 MB - Delta: 0.0 MB Correctness: 3/3 PASS — reasoning, religion, coding. Ceiling: 15.6 tok/s (on my single-GPU hardware). The architecture is model-agnostic. Any MoE. Any size! Shoutout to my dude 0xSero. We've been trading notes all week. He's got Kimi K2.5 running across 8×3090s! while we took different journeys on different hardware, we share the same obsession. Amazing collab. More soon..

need