Phone AI Assistant

After installing OpenWebUI via Docker on the Mac running the Ollama server, and switching remote access on the phone to TailScale, it reportedly became much more stable and more secure.

While on vacation in Hawaii, they used Gemma4:26b essentially like ChatGPT for everyday consultations, and found it fast, reliable, and favorable in terms of privacy.

Going forward, they plan to add SearXNG to bring in search and web crawling tools, which they believe would bring the local LLM closer to being an everyday assistant with real-time web access.

They also adjusted the response tendencies using OpenWebUI's system prompt to:

• reduce repetitive AI-style phrasing
• avoid excessive flattery
• reduce hallucinations
• collaboratively push back when gaps in reasoning appear

Not wanting boring, dry answers, they set the tone to be like a "slightly sarcastic nerd", and wrote that they are currently satisfied with that persona.