Troubleshooting
PAI won’t boot
Symptom: Machine boots into the host OS, ignores USB.
- Check your BIOS boot order — USB must be listed before the internal drive
- Try F12 (Dell/Lenovo), F9 (HP), Option (Mac Intel), or F8 (ASUS) at POST to get a one-time boot menu
- Verify the USB was flashed correctly — try re-flashing with
ddor Rufus - Disable Secure Boot in BIOS if enabled (UEFI → Security → Secure Boot: Disabled)
Symptom: GRUB loads but kernel panics on boot.
- Try the
nomodesetboot option: at GRUB, presse, addnomodesetto the kernel line, press F10
Ollama won’t start / no models available
# Check Ollama status
systemctl status ollama
# Restart Ollama
sudo systemctl restart ollama
# Pull a model manually
ollama pull llama3.2
# List available models
ollama list
If Ollama fails to start, check available RAM:
free -h
Models need free RAM to run:
- 3B models: ~2 GB RAM
- 7B models: ~5 GB RAM
- 13B models: ~10 GB RAM
Chat interface not loading
# Check Open WebUI service
systemctl status open-webui
# Restart Open WebUI
sudo systemctl restart open-webui
# Check if port 8080 is listening
ss -tlnp | grep 8080
If Firefox shows “Connection refused”, wait 30 seconds after boot for services to start, then refresh.
Wi-Fi not connecting
# List available networks
nmcli device wifi list
# Connect to a network
nmcli device wifi connect "NetworkName" password "password"
# Check network manager status
systemctl status NetworkManager
Tor (privacy mode) issues
# Enable Tor
sudo pai-privacy on
# Disable Tor
sudo pai-privacy off
# Check Tor status
systemctl status tor
# Verify Tor is routing (should show Tor exit node IP)
curl https://check.torproject.org/api/ip
Persistence not working
# Check persistence status
pai-persistence status
# Manual unlock (if auto-unlock failed)
sudo pai-persistence unlock
# View persistence service logs
journalctl -u pai-persistence
Display issues / blank screen
- Try a different output: most Sway configs use
*to auto-detect - Check if Wayland is running:
echo $WAYLAND_DISPLAY - For VM/software rendering issues, check environment variables:
export WLR_RENDERER=pixman export WLR_NO_HARDWARE_CURSORS=1 sway
Performance is very slow
CPU inference is slower than GPU inference. Expected speeds on 8 GB RAM laptop:
- 3B model: 15–40 tokens/sec
- 7B model: 5–15 tokens/sec
- 13B model: 2–5 tokens/sec
To improve performance:
- Use a smaller/quantized model (e.g.,
phi3:miniinstead ofllama3:8b) - Close other applications to free RAM
- Use a machine with AVX-512 CPU support
Getting help
- GitHub Issues — bug reports and feature requests
- README — full documentation