AI Mode
Automatic, reversible system tuning for local AI inference — the defining feature of Genesi OS.
AI Mode is a background daemon (genesi-aid) that detects when you're
running a local AI workload and tunes the system for inference — then restores
every setting when the workload ends. It's fully reversible: it captures
each value before changing it and puts it back on exit (and survives a daemon
restart via a /run snapshot).
What it tunes
| Setting | Normal | AI Mode |
|---|---|---|
| CPU governor | powersave | performance |
| Swappiness | 60 | 10 (keep weights in RAM) |
| Transparent huge pages | madvise | always |
| Inference process priority | 0 | −5 (higher) |
On machines with a GPU, AI Mode also detects available VRAM (via nvidia-smi,
DRM sysfs, or a vendor-agnostic Vulkan probe) to inform how much of a model can
be offloaded. The guiding principle is "runs great on a potato" — every knob is
RAM/VRAM-gated and degrades gracefully rather than pushing a weak machine into
OOM.
Supported runtimes
Ollama · llama.cpp (llama-server, llama-cli) · vLLM · LocalAI ·
text-generation-webui · KoboldCPP · Oobabooga.
How much faster?
Based on testing with Ollama on CPU-only systems:
- Tokens/second: +15–25%
- Model load time: −30–40%
- Memory stability: reduced swap usage
With a GPU the inference gains are smaller, but CPU/memory tuning still helps.
Using it
AI Mode activates automatically — just run a model. To control it manually,
use the genesi-ai-mode CLI:
genesi-ai-mode status # current state + detected processes
genesi-ai-mode on # force on
genesi-ai-mode off # force off
genesi-ai-mode auto # back to automatic detectionInspect the daemon directly:
sudo systemctl status genesi-aid
sudo journalctl -u genesi-aid -fAI Mode widget
The desktop ships a Plasma 6 widget (and a Monitor app) showing whether AI Mode is active, the detected processes, the applied optimizations, and live tokens/second.