How do you guys setup search with your AI models?

The discussion revolves around integrating the Gemma 4 model, specifically the 12 billion parameter variant, as a voice assistant with search capabilities. The user is currently utilizing a LlamaCPP container in a Docker Compose setup on Fedora Linux but is facing issues with search functionality using DuckDuckGo and prefers not to rely on external API keys from services like Brave or Google. This highlights the challenges of self-hosting AI models while maintaining effective search integration without third-party dependencies.