From paradise in Honolulu, another successful maintenance day on my local AI infrastructure.
I just finished a complete refresh of my Debian 13 (Trixie) AI workstation. The goal was simple: keep everything current, eliminate conflicts, and make future updates effortless.
What Got Updated:
- Ollama — Switched fully to the latest Docker version (native install disabled to avoid port conflicts). Models refreshed: llama3.2, deepseek-r1:14b, gemma3:12b, and qwen2.5:14b.
- Open WebUI — Pulled the latest image and restarted. Now talking smoothly to the new Ollama backend.
- Agent-Zero — Updated to the newest image and redeployed on port 5080.
- Deer-Flow stack (frontend, gateway, langgraph, nginx) — Rebuilt the custom images and resolved a stubborn Docker network overlap. Everything is now running on the fresh docker_deer-flow-dev network, with nginx proxy available at port 2026.
All containers are healthy, and I reclaimed over 20 GB of disk space during the cleanup.
The Best Part:
I created a simple one-line update command:
Bash
updateaiThis single alias now handles pulling the latest images, restarting services, rebuilding Deer-Flow when needed, and pruning old images. Future maintenance just got a lot smoother.
This setup brings me one step closer to a reliable, self-hosted JARVIS-like environment — running powerful local models with multiple agent frameworks, all on a stable Debian box in Honolulu.
Next steps on the roadmap: deeper integration between the tools, possible GPU acceleration tweaks (if hardware allows), and continued documentation of the journey.
Stay tuned for more JARVIS Intelligence from Paradise.
— Dennis (@ny2honolulu)