mac-studio

Autonomous Personal AI Assistant for office work, exposed as a Model Context Protocol (MCP) server and listed on the Official MCP Registry as io.github.Graphene-Lab/agentbridge. Local-first: drives tools, mail, documents and voice on your machine.

M5 Ultra vs M3 Ultra Mac Studio at local AI: every test, run, result and explanation (dense vs MoE, why reading got ~4x and writing ~1.5x). Raw data + the harness to rerun it.

Advanced SSD prefill cache for Mac — MLX, exo

NotchX is adds MacBook like notch to Windows ,Linux and Mac Desktop devices at the same time providing better user experience

MLX fine-tuning toolkit for Mac Studio M3 Ultra 512 GB — distill Claude Opus reasoning into Mistral Large 123B for the ailiance LLM stack.

MagicMirror² module to monitor an Ollama/Hermes LLM host (Mac Studio): GPU, memory, loaded models & VRAM, over MQTT.

Private, self-hosted AI office assistant for Apple silicon Macs (Mac Studio M5 Max/Ultra, MacBook Pro, MacBook Air, iMac, Mac mini, Mac Pro): local documents, DOCX/XLSX/PPTX/PDF, email, Telegram, voice — fully offline with a local model (Ollama).

Superseded as the default by minimax-m31-flash-exl3-m5-ultra (closer to the release). MiniMax-M3.1-Flash as MLX conversions on one M5 Ultra, TensorFold native Metal, decode ~43 tok/s

Sovereign local LLM hardware guide (mid-2026): memory bandwidth as the real bottleneck, frontier open models, Unsloth/MLX quantization, Thunderbolt 5 clusters, Mac M5 Ultra vs DGX Spark vs AMD vs RTX 6000 Pro. Bilingual FR/EN. CC BY-NC-ND 4.0.

Lab notebook of a German craftsman running his solar business on local AI: 2x DGX Spark, Mac Studio, MCDMA. Recipes, benchmarks, mistakes. Every source credited.

Benchmark harness for local LLMs on Apple silicon. One llama.cpp cell, every condition the number depended on, as JSON. AI-operated.

Apple 芯片 Mac mini 与 Mac Studio 风扇监控和控制工具

Benchmarks and serving notes for Qwen3.5-27B fine-tune on Apple Silicon via MLX. Companion to dgx-spark-nvfp4-serving.

ローカルLLMを動かすPCの機種比較。購入できる構成だけで統合メモリ×ストレージを並べた調査資料

不踩坑：GLM-5.3-Flash (Mac oMLX) — Jinja output discipline + OpenAI-compatible proxy + caller-side validation. Three-layer defense for stable GLM output on Mac Studio.

macOS menu bar app for headless Mac mini / Mac Studio home servers — live CPU, RAM, SSH & port status over LAN, Tailscale & Thunderbolt, plus one-click SSH & SMB

Measured (not estimated) local-LLM benchmarks on a 64GB Mac Studio M4 Max: DwarfStar ds4-bench, 2K–64K context, raw logs + reproduce script

DeepSeek-V4.1-Flash-1017 on a Mac Studio M5 Ultra: bit-identical decode speedups (15.6 → 14.6 ms a token) for TensorFold's native Metal engine, with profiling tools and an exactness gate

Run local LLMs on your Mac and use it from anywhere — a macOS AI assistant + native RDP server. Ollama & MLX, OpenAI/Anthropic-compatible API, on-device & private. Apple Silicon.

Open benchmark data for local LLMs on Apple silicon: what fits in 512 GB, and predictions published before the machine shipped. CC BY 4.0.