Observer AI adds parameter tuning, Gemma 4 reasoning

Observer AI is positioning itself as the privacy-first alternative to commercial "Recall" features by giving developers granular control over their local agent's inference. Parameter tuning (Temp, Top P, Max Tokens) enables high-reliability zero-shot detection for real-time visual triggers. Native support for Gemma 4's "Reasoning Effort" toggle allows users to prioritize either latency or deep thinking depending on the monitoring task. Custom API key support makes the tool a versatile BYOK platform for both local Ollama setups and remote hosted models. The 100% local architecture directly addresses the massive privacy concerns surrounding persistent AI-driven screen capture and activity logging.