Four Failed Attempts Before Trading Sidecar Finally Clicked
I've tried building something like this at least four or five times. I love to trade and wanted AI to help me analyze charts in real time, suggesting trade ideas, profit targets, and when to move stop losses. Every previous attempt hit a wall. This time, the approach finally clicked.
Four Failed Attempts
The first attempt used raw stock data APIs. The pure numerical data wasn't what AI needed to understand chart patterns. It couldn't "see" the chart. Next I tried a Chrome extension to take screenshots of TradingView, but browser security sandboxing made it unreliable. Then I built a Node.js instance on a PC that would screenshot the session, send it to Firebase, and have a React app check for new images and generate a response. It worked, technically, but the round trip was so slow it wasn't faster than just manually screenshotting and uploading to Claude in the browser.
Then it hit me: I've been building native iOS apps with Claude Code without issue. Why not build a native macOS app that does this directly? No browser extensions, no Node servers, no Firebase round trips. Just a native app with a built-in web view and an AI sidebar.
The Breakthrough
In a single weekend, I had a working prototype. A macOS app with a main web view supporting multiple browser tabs for different charts and brokerages, custom prompt creation, configurable screenshot capture intervals, and a suggestions engine that analyzes the active tabs and generates trading insights.
Evolving the Suggestion Engine
Three iterations, each informed by real trading sessions
v1: Plain Text
The first version returned raw text analysis. Useful information, but hard to scan quickly when markets are moving. Trading requires fast decision-making and walls of text don't support that.
v2: Action Cards
Restructured into card-based suggestions with clear actions: OPEN LONG, STAY OUT, EXIT. Each card shows the ticker, P&L, entry/current/stop-loss prices, and a specific recommended action like "Move Stop to 173.70." Color-coded for instant parsing: green for active positions, amber for caution.
v3: Strategy Cards
The current iteration goes beyond single-action suggestions. AI now generates scenario-based strategy cards that outline a range of possibilities. For example: "Watch for: break above 276 MA Ribbon resistance with volume for long, or breakdown below 274 support for short." This gives precise price levels and conditions to watch for, letting you prepare for multiple outcomes instead of reacting to one prediction.
Other Features
App Settings
Configurable appearance, layout, and behavior. Theme support (system, light, dark), adjustable chat text size, flexible panel positioning, screenshot quality settings, custom prompts, auto analysis intervals, and API configuration.
Tab and Multi-Chart Layout
The top bar with tabs mirrors a browser but with trading-specific affordances. The web view supports multiple tabs so you can have different charts and brokerages open simultaneously, and the app intelligently captures screenshots from all tabs or specific ones.
Designed in Figma, Built with AI
This is the most complex AI integration of any of my side projects, and the difference matters: I didn't just use AI to build the app, the app itself is an AI product. The actual design problem is how a human and an AI collaborate on real-time decisions that have real money attached.
Every design iteration came from actual use. The evolution from text to action cards to strategy cards was driven by trading with the tool and finding the gaps. That tight loop between designing, building, and using the product myself is what makes these side projects valuable: the feedback is immediate and honest.