Rezon is an intelligent desktop overlay that learns your trading style and reacts to your actions in real time. It observes how you trade—what you hover, click, and explore—and provides adaptive insights, warnings, and recommendations using a locally hosted AI agent built on Qwen2.
Unlike traditional trading tools that only show data, Rezon helps you understand your own behavior as a trader.

• Live Browser Integration: Connects to your trading or finance site via Chrome Remote Debugging to interpret hovers, clicks, and interactions.
• Personalized AI Agent: Uses a local Qwen2 model to analyze behavior and tailor feedback to your trading style.
• Autonomous Tool Use: The agent dynamically decides when to pull live stock data through YFinance or relevant market insights from DuckDuckGo.
• Privacy-Preserving: All AI reasoning happens locally—no cloud inference, no user data collection.
• Session Summaries: Automatically generates a reflective report at the end of each session with insights and behavioral trends.

Most trading platforms focus on automation and data visualization but neglect the human element behind every decision. Traders often fall into recurring habits—selling too early, holding risky stocks too long, or reacting emotionally to market noise.
Rezon was built to bridge that gap: an intelligent overlay that understands your unique trading psychology and helps you make better decisions through real-time reflection.

1. Personalization Setup:
   When you start Rezon, you describe your trading personality—your goals, tendencies, and weaknesses (like "I hold volatile stocks too long" or "I sell too early").

2. Browser Monitoring:
   The Electron overlay attaches to your Chrome browser via the Chrome Remote Debugging Protocol and listens for meaningful DOM updates and hover events.

3. Agentic Node Reasoning:
   The local Qwen2 model runs within a Python backend, organized as a directed graph of agentic nodes. Each node represents a reasoning or action step.
   Based on your context, it autonomously decides whether to:

   • Fetch live stock prices from YFinance
   • Retrieve news or market context using DuckDuckGo
   • Generate personalized feedback or warnings
   • Summarize reasoning for the user

4. Session Summary:
   When you finish trading, Rezon compiles your behavioral log, AI responses, and trade outcomes into a concise end-of-session summary that highlights improvement areas and recurring patterns.

The Electron frontend handles the user interface and overlay rendering, while a WebSocket connection bridges Chrome's debugging layer and the Python backend.
Every significant browser interaction, like viewing a chart, hovering over a "Buy" button, or opening a stock preview, is converted into a structured event.

These events feed into a Python-based agent graph, where Qwen2 serves as the reasoning core. Each “node” in the graph can analyze the situation, make decisions, or call tools. This modular setup makes it easy to expand with new reasoning steps or data sources.

By keeping the model and tools local, we achieve near-instant feedback while preserving full user privacy.

• Maintaining smooth browser performance while streaming DOM updates.
• Synchronizing real-time events between Electron, Chrome, and the Python agent layer.
• Designing an efficient event schema for Qwen2 that captures useful behavior without unnecessary noise.
• Optimizing local model performance while maintaining reasoning quality.

• Built a functioning, privacy-preserving AI trading companion that runs entirely locally.
• Designed an agentic reasoning system capable of autonomous tool use and real-time feedback.
• Integrated live data sources like YFinance and DuckDuckGo for accurate, contextual AI decisions.
• Created a dynamic event model that bridges browser behavior with structured AI reasoning.

• Real-time personalization significantly enhances the relevance of AI feedback in trading.
• Local models like Qwen2 can perform complex reasoning with minimal latency when used efficiently.
• The key to user trust lies in balancing assistance with transparency. Users want guidance, not surveillance.
• Building agentic node graphs for LLMs enables modular, traceable decision-making systems.

• Integrate with brokerage APIs for real trade tracking and richer execution context.
• Expand the agentic graph with deeper reasoning nodes for risk analysis and market prediction.
• Add intuitive visual overlays for feedback and alerts within trading UIs.
• Extend support to new markets such as cryptocurrency and forex.