Melissa

Project Melissa: Local Autonomous AI Assistant ​Melissa is a custom, web-based autonomous AI software agent and user interface engineered to run completely locally on mobile hardware configurations. Designed to bypass the privacy concerns and subscription costs of cloud-reliant platforms, Melissa brings robust, context-aware analysis directly onto the device. ​🛠️ Core Architecture & Backend Engineering ​Localized Architecture: Powered by a lightweight Python Flask web server integrated directly with a llama.cpp inference server. ​Self-Contained Environment: Optimized to compile from source and execute entirely within an Android Termux environment, operating fully independent of external servers. ​Sliding Window Memory: Implements a custom orchestration layer that manages a rolling history of the most recent conversation turns. This prevents token overflow, respects hardware RAM limitations, and eliminates the "context amnesia" typical of small-scale local models. ​Direct Math Evaluation Bypass: Features an intelligent regex-driven interceptor that handles straightforward arithmetic operations natively within the Python backend. This guarantees instant, hallucination-free calculations while preserving processing cycles on the local LLM server. ​Hardware Spike Patience: Configured with an extended 120-second connection timeout threshold to gracefully handle occasional generation spikes inherent to local mobile processor inference. ​📱 Premium Frontend & User Experience ​Optimized Mobile Shell: Features a highly responsive, modern dark-mode interface designed specifically for mobile screen real estate and rapid beta-testing workflows. ​Session History Sidebar: Incorporates a slide-out navigation menu displaying active session data, equipped with real-time local text search filtering to locate past inputs instantly.