🔐 This page is for system administrators only. All commands below require SSH access to the server. If you're an end user looking for help, visit Troubleshooting or FAQ instead.
Quick reference — most common admin tasks
Restart all services
sudo systemctl restart ollama open-webui chromadb jupyter
Check all service status
sudo systemctl status ollama open-webui chromadb jupyter
Check disk space
df -h /
Check GPU status
nvidia-smi
List installed models
ollama list
View live system logs
sudo journalctl -f -u ollama -u open-webui
Service management
All Nexus One AI services run as systemd units. Replace SERVICE below with: ollama, open-webui, chromadb, or jupyter.
Start a service
Use when a service is stopped and needs to be brought up.
sudo systemctl start SERVICE
Stop a service
Gracefully stops the service. Active users will be disconnected.
sudo systemctl stop SERVICE
Restart a service
Stop then start. Use when a service is misbehaving or after a config change.
sudo systemctl restart SERVICE
Check service status
Shows whether the service is running, the last few log lines, and any errors.
sudo systemctl status SERVICE
Enable auto-start on boot
Ensures the service starts automatically if the server is rebooted.
sudo systemctl enable SERVICE
Restart all Cezen services
Restart everything at once. Useful after a server update or if multiple services are unresponsive.
sudo systemctl restart ollama open-webui chromadb jupyter
Service ports and endpoints
| Service | Port | URL | Notes |
|---|---|---|---|
| Open WebUI | 3000 | http://ai.local:3001 | Main user-facing chat interface |
| Ollama API | 11434 | http://ai.local:11434 | Model inference API — do not expose externally |
| ChromaDB | 8000 | http://ai.local:8000 | Vector database API — do not expose externally |
| Jupyter | 8888 | http://ai.local:8888 | Notebook server — restrict to developer users only |
| FastAPI | 8080 | http://ai.local:8080 | Custom API layer — if deployed |
Model management
List installed models
Shows all models currently available on the system with their sizes.
ollama list
Pull (install) a new model
Downloads and installs a model. Examples:
llama3.1:8b, mistral:7b, gemma2:9b. Requires the model files to be available on your network or internet (if connected).ollama pull MODEL_NAME
Remove a model
Deletes a model from disk to free up space. Make sure no users are actively using it.
ollama rm MODEL_NAME
Show model info
Displays the model's parameters, context length, and configuration.
ollama show MODEL_NAME
Check what's currently loaded in GPU
Shows running Ollama processes and their VRAM usage.
ollama ps
Model storage location
All model files are stored here. Monitor this directory for disk usage.
ls -lh /usr/share/ollama/.ollama/models/
User management (Open WebUI)
User accounts are managed through the Open WebUI admin panel at http://ai.local:3001/admin/users. Log in with your admin account and use the panel for most user tasks. The commands below are for bulk operations or when the UI is unavailable.
Open user admin panel
The primary way to create, disable, and manage user accounts. Open this URL in a browser while connected to the internal network.
http://ai.local:3001/admin/users
Open WebUI database location
User data is stored in a SQLite database at this path. Back this up regularly.
/opt/open-webui/data/webui.db
Back up Nexus One AI data
Creates a local archive of Cezen data, settings, audit logs, and uploaded working files.
sudo bash /opt/cezen/scripts/cezen-backup.sh backup
System health & monitoring
GPU status snapshot
Shows GPU utilisation, VRAM used/free, temperature, and power draw for all GPUs.
nvidia-smi
Live GPU monitor
Refreshes every second. Good for watching GPU load while a model is running. Press Ctrl+C to exit.
watch -n1 nvidia-smi
Full GPU health diagnostic
Runs DCGM's built-in diagnostic suite. Takes ~2 minutes. Red results indicate hardware issues to escalate.
dcgmi diag -r 1
Disk space usage
Check overall disk space. Watch /: models, databases, and logs all grow over time.
df -h
Find what's using the most disk space
Shows the 10 largest directories under /opt — useful when disk space runs low.
sudo du -sh /opt/* | sort -rh | head -10
CPU and memory overview
Live system resource usage — CPU load, RAM, swap. Press Q to exit.
htop
Check server uptime
Shows how long the server has been running and current load averages.
uptime
Log access
Live logs — all Cezen services
Streams live log output from Ollama and Open WebUI together. Useful for diagnosing live issues. Press Ctrl+C to exit.
sudo journalctl -f -u ollama -u open-webui
Last 100 lines — Ollama
Recent Ollama log output. Check here first when models aren't loading or inference is failing.
sudo journalctl -u ollama -n 100 --no-pager
Last 100 lines — Open WebUI
Recent Open WebUI log output. Check here when users can't log in or the interface is not loading.
sudo journalctl -u open-webui -n 100 --no-pager
Logs since last boot
All logs from the current boot session — useful after a server restart to check everything came up cleanly.
sudo journalctl -b -u ollama -u open-webui -u chromadb -u jupyter
Search logs for errors
Filters log output to show only error-level entries — faster than reading everything.
sudo journalctl -u ollama -u open-webui -p err --no-pager
Common procedures
Safe server reboot
Before rebooting, notify users and allow active sessions to finish. All systemd services with enabled status will restart automatically on boot.
- Notify users that the system will restart in 10 minutes
- Wait for active model requests to complete: watch ollama ps
- Reboot the server: sudo reboot
- After reboot, verify all services are running: sudo systemctl status ollama open-webui chromadb jupyter
- Open http://ai.local:3001 and confirm the chat interface loads
Adding a new AI model
Models are added via the Ollama CLI. On an air-gapped system, copy the model file to the server first.
- Check available disk space: df -h / — ensure at least 2× the model size is free
- Pull the model (if internet-connected): ollama pull llama3.1:8b
- For air-gapped install, copy the GGUF file then: ollama create mymodel -f /path/to/Modelfile
- Verify it appears: ollama list
- Test it loads: ollama run MODEL_NAME "Hello"
- The model will appear automatically in Open WebUI's model selector — no restart needed
Freeing up disk space
Disk space is most commonly consumed by model files, ChromaDB data, and Jupyter notebooks. Check each before deleting anything.
- Find what's using the most space: sudo du -sh /opt/* | sort -rh | head -10
- List models with sizes: ollama list
- Remove unused models (confirm with users first): ollama rm MODEL_NAME
- Clear old journal logs (keeps last 7 days): sudo journalctl --vacuum-time=7d
- Verify space recovered: df -h /
Resetting an admin password (Open WebUI)
If the admin password is lost, reset it via the Open WebUI CLI while the service is stopped.
- Stop Open WebUI: sudo systemctl stop open-webui
- Run the reset command: sudo open-webui reset-admin-password --email admin@example.com --password newpassword
- Restart Open WebUI: sudo systemctl start open-webui
- Log in with the new password and immediately change it to something secure
Need Cezen support?
For hardware issues, system-level failures, or anything beyond this guide, contact Cezen. Include: the service name, the error from journalctl, and the output of nvidia-smi and df -h.