The user runs a self‑hosted personal AI agent (OpenClaw) 24/7 on an M4 Mac mini, relying on cloud‑based Claude Opus as the main model with GPT and DeepSeek fallbacks, costing about $400 per month. They are asking whether a Mac Studio equipped with an M5 Ultra could host the entire workflow locally, eliminating the need for any cloud subscriptions. The agent currently performs long, multi‑step conversations that involve tool calls such as reading files, running scripts, and searching email and calendar data.
Read original
reddit/r/LocalLLM