APEX: paper trading workspace

A desktop app for practising trading strategies with fake money on real market prices. It runs on macOS and Windows.

• Live prices: Binance for crypto, Yahoo Finance for forex, gold, stocks and indices
• Trading chart with candles, EMA overlays, your positions, and stop loss / take profit lines you can drag
• TradingView view, plus a button that opens the symbol on tradingview.com
• Paper broker that behaves like a real margin account (details below): bid/ask fills, fees, leverage limits, margin close-out, market hours
• Orders: market, limit, stop and stop limit, with stop loss / take profit, close one or close all, account reset
• Risk limits that actually get enforced: risk per trade, max drawdown, daily loss limit, max open positions
• 10 working strategies, each with real signal code, editable parameters, and your own copies
• Backtester for every strategy: same costs and fill rules as the paper account, with equity curve, drawdown, profit factor, Sharpe and a trade list
• Three modes: Manual, AI Setup (AI tunes the selected strategy and shows where it stands), AI Auto (AI tunes it, then the engine trades its signals for you)
• AI chat that knows your selected strategy and account (optional, any OpenAI-compatible API such as FreeLLMAPI)

Nothing here touches a real broker. Every fill is simulated on your own computer.

How the simulation works

It's built to behave like a European retail CFD / margin broker, so you don't get used to things a real account won't let you do.

Market	Max leverage (ESMA)	Spread used when the feed has none	Commission
Crypto	2:1	real Binance bid/ask	0.10% per side
Forex	30:1	about 1 bp	none (spread only)
Gold	20:1	about 1.4 bp	none
Indices	20:1	about 1.2 bp	none
Stocks	5:1	about 2 bp	0.02%, $1 minimum

• Margin. Every position ties up size x price / leverage. New trades need enough free margin. If equity falls under 50% of used margin, everything gets closed out, like a real broker would.
• Fills. Market buys fill at the ask and sells at the bid, plus 1 bp of slippage. Limit orders and take profits fill at your price. Stops fill at the stop, or worse if price gaps past it.
• Nothing gets skipped. Every 15 seconds the engine checks the 1-minute candles since the last check, so a spike through your stop between checks still fills.
• Market hours. Crypto trades 24/7. Forex and gold run Sunday 17:00 to Friday 17:00 New York time. US stocks and indices run 09:30 to 16:00 New York, the DAX 09:00 to 17:30 Frankfurt. When a market is closed, market orders are refused and working orders wait for the open. Holidays aren't modelled.
• Strategies decide on a finished candle and act on the next one. The backtester and AI Auto use the exact same code, so a backtest shows how the bot would really have traded.
• Not modelled: overnight financing (swap), partial fills, market impact for large orders, exchange holidays.

How it's built

Part	Tech	Folder
Desktop shell	Tauri 2 (Rust)	src-tauri/
UI	React + TypeScript + Vite, charts by lightweight-charts	src/
Backend	Python FastAPI, bundled as a sidecar with PyInstaller	backend/

The Tauri shell owns the Python backend:

• It starts it on a random local port with a one-time token and waits until it answers. If it can't start, the app shows the real error and a Try again button instead of a dead window.
• The UI never talks HTTP itself. Every call goes through a Rust command (api in src-tauri/src/main.rs), so macOS WebKit can't block it.
• The backend runs in its own process group (macOS) / job object (Windows). PyInstaller's one-file build is two processes, and quitting or updating now kills both. The backend also exits by itself if the app is force-quit or crashes (--parent-pid), and the Windows installer stops any leftover apex-backend.exe before copying files (src-tauri/windows/hooks.nsh).
• Updates download in the Rust shell, which stops the backend right before the installer runs.

Your data (paper account, settings, logs) lives here:

• macOS: ~/Library/Application Support/APEX/
• Windows: %APPDATA%\APEX\

What you need

• Node.js 22 (20 and 24 work too): https://nodejs.org
• Python 3.10 or newer: https://www.python.org. On Windows, tick Add python.exe to PATH in the installer.
• Rust: https://rustup.rs
• macOS only: xcode-select --install
• Windows only: Desktop development with C++ from the Visual Studio Build Tools installer (rustup offers to install it). WebView2 already ships with Windows 10 and 11.

Moving the folder between Mac and Windows? Delete node_modules, .venv-backend and src-tauri/target first. They only work on the OS they were made on. Everything gets recreated on the next install or build.

Build the app

The same commands work in Terminal (macOS) and PowerShell (Windows):

npm install
npm run build:app

That builds the backend, the UI and the desktop app, and drops the result in release/:

• macOS: APEX.app and a .dmg
• Windows: a -setup.exe installer and an .msi

macOS: if DMG packaging fails with hdiutil ... Resource busy, run npm run build:app -- --no-dmg to build just the .app.

Releases and updates

Installers are built by GitHub, not on your own computer, so the Mac and Windows versions always come out of the same code.

• Ship a version: on GitHub open Actions > Release > Run workflow, type the next version (like 0.3.0) and one or two sentences of what's new, then Run workflow. In about 15 minutes there's a release with APEX-Setup.exe (Windows) and APEX-mac-arm64.dmg (Apple Silicon Mac).
• Test build: same button with the version left empty. The installers show up under Artifacts on the run and nothing is published.
• Auto updates: the app checks GitHub a few seconds after launch and every 6 hours. When there's a new version a card pops up with the notes and an Update and restart button. Settings also has Check for updates.
• Updates are signed. The private key is the TAURI_SIGNING_PRIVATE_KEY repo secret (Settings > Secrets and variables > Actions). Keep a backup of it somewhere safe: if it's lost, installed apps can't take new updates and everyone has to reinstall by hand once.
• Every push and pull request runs CI: typecheck, UI build, a backend start test and a Rust compile check.

Work on it

Fast loop in the browser (no Rust needed). Use two terminals:

npm run dev:backend   # Python backend on port 8765
npm run dev           # UI on http://localhost:1420

Inside the real desktop window:

npm run backend       # build the Python sidecar once (again whenever backend/ changes)
npm run tauri dev

AI setup (optional)

Click ⚙ in the top right.

• Base URL: http://127.0.0.1:31415/v1 for FreeLLMAPI, or any OpenAI-compatible endpoint
• Model: auto if your gateway routes automatically, otherwise a model ID
• API key: your key or FreeLLMAPI Superkey

Hit Save and test. The key is stored in config/openai.json inside the data folder above and is never sent back to the UI.

Limits worth knowing

• Yahoo prices can be delayed, and the app labels them that way. Binance spot prices are close to real time.
• Yahoo has no 4h candles, so APEX builds them from 1h candles.
• The engine (fills, stops, AI Auto) runs inside the app's backend. The app has to stay open with the engine running. After a restart the engine always starts paused, on purpose.
• Some strategies need interpretation of their spec, e.g. Swing MTF treats 4 chart candles as one higher-timeframe candle, and the scalping spread filter is in basis points.
• The TradingView view is display only. TradingView has no public API for third-party apps to place trades in your TradingView paper account, so the two accounts are separate.
• Backtests and AI suggestions are research, not predictions.