Recent events, including Claude Code restricting access in China and a Chinese government security bulletin on AI coding tools, have renewed concerns about where code data goes when processed by AI as…
An independent benchmark of on-device LLMs on iPhone A17 Pro found Apple's system model achieves ~149 tok/s with only 12MB peak app memory, while 4B-class open models like Qwen3 4B and Llama 3.2 3B tr…
Ollama 0.31 introduces multi-token prediction for Gemma 4 on Apple Silicon, cutting token generation time by 90% on Aider's coding benchmark. The update automatically accelerates local coding agents w…
A local AI assistant called Kira discovered that nearly half of its model time was spent re-reading boilerplate prompt text—a 'prefill tax'—after migrating from Ollama to Apple's MLX framework, which …
Qwen open-sourced the 35-billion parameter Mixture of Experts model Qwen 3.6-35B-A3B, which activates only 3 billion parameters per token and runs on a $599 Mac Mini M4 with 16GB RAM at 17 tok/s with …
A developer released mlx-serve, a native Zig server for MLX-format language models on Apple Silicon, enabling local, free, and private use of AI coding assistants like Claude Code. The server exposes …
Ollama's new MLX engine enables local LLM inference on MacBook Air at twice the speed, making powerful AI accessible on consumer hardware without cloud dependency.…
Uber capped employee AI coding tool spending at $1,500 per month in June 2026 after burning through its entire 2026 budget in four months, with Claude Code adoption surging from a third of engineers t…
The Mano-P team released Mano-CUA 2.0, a 4B-parameter GUI agent that runs on a MacBook, and found that the primary bottleneck was not model size but insufficient Chinese GUI training data. After addin…
William, a tiny poetry language model trained by a developer, runs entirely in the browser using ONNX Runtime Web. The 6-layer transformer was trained on the Gutenberg Poetry Corpus and fine-tuned on …
A developer built Squish, a local LLM inference server for Apple Silicon, after finding existing tools like Ollama too slow for generating git commit messages. Squish achieves up to 9.8× faster perfor…
Gingerpaw, a new macOS app, launches a multi-agent coding workspace with on-device voice dictation and a talking ginger cat notification system. The app runs multiple coding agents like Claude Code an…
A new Local LLM Hardware Calculator helps users estimate memory requirements for running large language models on their own hardware, factoring in weights, KV cache, and overhead. The tool also compar…
Alibaba's Qwen team released Qwen3-30B-A3B, a Mixture-of-Experts model with 30.5 billion total parameters but only 3.3 billion active per token, enabling it to run on a single 24 GB graphics card at s…
Co/Core launches an AI cooperative where members share their own compute hardware to run AI inference jobs, using an open standard and public receipts to ensure transparency. The platform supports exi…
Apple showcased its latest AI tools for developers during a 90-minute WWDC26 presentation at the Steve Jobs Theater, including a live demo of an entire app built from a single prompt. The session high…
Researchers reproduced the core mechanisms of two 2025-26 JEPA papers (CrossJEPA and LeJEPA) at toy scale on an M5 MacBook Air, confirming that the self-supervised representation learning methods work…
A researcher using nanoGPT on a MacBook Pro faces dependency failures when moving to a GPU cluster, prompting a solution using Nix and Flox to create reproducible, cross-platform runtime environments …
Thunder Kitty 1.9.0 introduces live topic segmentation and agenda tracking that run entirely on-device on Apple Silicon, using three models simultaneously. The developer behind the app spent seven att…
QuantaMind has developed a platform that addresses fragmentation in local LLM tool-calling by using a text-prompt structure as a universal baseline and providing a side-by-side native function-calling…