Releases·Tool Release
Developer ships TUFF 3.0, a macOS app that runs MoE LLMs larger than RAM via SSD streaming
A solo developer released TUFF 3.0, a native Apple Silicon macOS app that runs mixture-of-experts LLMs too large for system RAM. It keeps shared MoE components resident and streams needed experts from SSD through a bounded cache, letting a 61GB model run on a 16GB MacBook Air without swapping. The app ships with five built-in models, a chat UI, and a local OpenAI-compatible server, making large local inference practical on low-memory Macs.
7 engagement1 sourceSun, Aug 30, 2026, 08:22 PM
AI Overview
- Keeps shared parts of a mixture-of-experts model resident in RAM and streams required experts from SSD through a bounded cache
- Runs a 61GB model on a 16GB MacBook Air without swapping
- Includes five built-in models: Gemma 4 variants, Qwen3.6, and GPT-OSS 20B/120B
- Chat UI supports Markdown, LaTeX, and file attachments
- Exposes a local OpenAI-compatible server; app is fully native on Apple Silicon
- Started as a fork of a small proof-of-concept and grew into a full app built by one developer
Related
Releases·Tool Release
TurboFieldfare runs Gemma 4 26B-A4B in ~2 GB RAM on Apple Silicon Macs
A developer released TurboFieldfare, a custom Swift + Metal runtime that runs the instruction-tuned Gemma 4 26B-A4B model in about 2 GB of RAM on any M-series MacBook, including 8 GB models. This enables local inference of a 26-billion-parameter model on memory-constrained devices.
A developer released TurboFieldfare, a custom Swift + Metal runtime that runs the instruction-tuned Gemma 4 26B-A4B model in about 2 GB of RAM on any M-series MacBook, including 8 GB models. This enables local inference of a 26-billion-parameter model on memory-constrained devices.
63 engagement1 sourcegithubFri, Jul 17, 2026, 03:57 PM
Products·launched
Pocket AI Lab - open source iOS app for running LLMs fully on-device
Pocket AI Lab is a free, MIT-licensed iOS app that runs LLMs entirely on the iPhone with no cloud or account. Users paste a Hugging Face model link and the app detects the format to start chatting. It targets people who want to quickly try newly released open models locally on-device.
Pocket AI Lab is a free, MIT-licensed iOS app that runs LLMs entirely on the iPhone with no cloud or account. Users paste a Hugging Face model link and the app detects the format to start chatting. It targets people who want to quickly try newly released open models locally on-device.
11 engagement1 sourceredditThu, Sep 10, 2026, 11:00 AM
Products·launched
Ghost: Mac app for reliable multi-step work with local Ollama models
Ghost is a Mac app that turns Ollama into a tool harness for real work, not just chat. It calibrates models to handle tool calling reliably, routing around flaky JSON output, and wraps the harness in a chat interface.
Ghost is a Mac app that turns Ollama into a tool harness for real work, not just chat. It calibrates models to handle tool calling reliably, routing around flaky JSON output, and wraps the harness in a chat interface.
2 engagement1 sourceredditFri, Aug 14, 2026, 08:19 PM
Releases·Tool Release
Samosa Chat lets users run Qwen3.6-35B-A3B locally on 16 GB Mac
A developer released Samosa Chat, a tool that enables running the Qwen3.6-35B-A3B model locally on a 16 GB Mac. The project is hosted on GitHub and was shared on Hacker News on July 15, 2026.
A developer released Samosa Chat, a tool that enables running the Qwen3.6-35B-A3B model locally on a 16 GB Mac. The project is hosted on GitHub and was shared on Hacker News on July 15, 2026.
15 engagement1 sourcehackernewsWed, Jul 15, 2026, 02:31 AM