Mac Studio user vibe-codes a local AI cluster monitor app in hours, prompt included
SquirrelEStuff · reddit · 2026-09-06
Running two TB5-linked Mac Studios (512GB + 256GB), the author was tired of juggling Exos for RAM stats and screen-sharing into oMLX to check loaded models. Using Hermes and GLM-5.3-flash as coding agents, he built a native iOS/macOS cluster dashboard in a few hours:
- One card per compute node with live RAM, disk, and installed/loaded AI models
- Load/unload models directly from the app
- Home-screen widgets plus Siri queries about loaded models and usage
- He shares the full self-contained prompt: the agent must first send a setup questionnaire (node inventory, inference servers like oMLX/LM Studio/Ollama/exo, auth mode, telemetry source) before implementing, avoid hardcoded environment specifics, and never run destructive operations without permission.
More from coding & agent
- Same model, 63% vs 97% on ARC-AGI-3: 7 harness rules beat max reasoning — victor_explore · 2026-09-07
- Matt Shumer: I 4x'd my Pro/Max subs as agentic models remove the attention bottleneck — mattshumer_ · 2026-09-07
- GPT-6 Astra codes the geometry, Dreamina Seedance 2.5 renders the final footage — HeyAmit_ · 2026-09-07
- Grok Bot's topic-based agent form factor hailed as better than threads — brandon_galang · 2026-09-07
- Codex defaults to 3 subagents — one config.toml line raises concurrency to 64 — daniel_mac8 · 2026-09-07
- Teaching Qwen Next 3D sculpting in Blender by having GPT Astra coach it via MCP — LegacyRemaster · 2026-09-07