Local AI GPU Guard for n8n + Ollama: 4 Presets That Stop VRAM Crashes

Four ready-to-import n8n workflows that unload idle Ollama models on your schedule, plus a fake Ollama server so you can test with no GPU, and a log summarizer. Dry-run by default. MIT.

The free template unloads idle Ollama models. This pack makes it fit your machine:

- 4 presets (standard, gaming PC, shared workstation, always-on server). Same approved logic, tuned settings.

- A mock Ollama server for safe testing without a GPU

- An offline decision checker and a per-day log summary

- Every preset ships in dry-run mode, with no credentials

Honest limits: model-level only (it does not read VRAM), estimates idle time from Ollama's expires_at, and is for self-hosted n8n only.

Built by a retired military veteran with AI coding assistance. Tested with scripted checks; the list is in SETUP.md.