never-miss

A cross-platform config-driven ticket-purchasing bot for macOS and Windows. Connects to a live Chrome browser via CDP, extracts interactable elements from the DOM, walks a platform-specific FSM, and matches elements via a hybrid Aho-Corasick + Sentence-BERT pipeline.

Login is performed manually by the user before starting the bot.

Requirements

• macOS (Intel or Apple Silicon) OR Windows 10/11
• Rust 1.70+
• Google Chrome
• Python 3.10+ (for model export scripts only)
• ONNX models in models/ directory (see ARCHITECTURE.md for setup)

Hardware Acceleration: NeverMiss automatically detects your OS and hardware to prioritize the optimal execution provider for the ONNX models:

• macOS: CoreML (Apple Neural Engine / Silicon) → CPU fallback.
• Windows: TensorRT / CUDA (NVIDIA GPUs) → DirectML (AMD/Intel GPUs) → CPU fallback.
• Linux: TensorRT / CUDA → CPU fallback.

Note: The bot was developed on macOS, so the .dmg installation file can be found directly in the releases. Technically, it can run on Windows, but users need to train the models and compile them themselves.

Building and Packaging

This project can be easily compiled and distributed as a standalone app/folder for macOS and Windows.

For macOS

Run the included bash script to compile the release and construct a native macOS .app bundle and .dmg disk image:

./pack_mac.sh

This script will:

1. Compile the Rust binaries natively.
2. Generate the Apple AppIcon.icns for the squircle dock icon.
3. Bundle the models/ and config.yaml directly into NeverMiss.app/Contents/Resources/.
4. Output a ready-to-distribute ~/Downloads/NeverMiss.dmg with a drag-and-drop /Applications folder shortcut.

For Windows

Run the included PowerShell script to compile the release and gather all necessary files into a portable folder:

.\pack_windows.ps1

This creates a NeverMiss-Windows\ folder containing never-miss-gui.exe, the core bot executable, ONNX DLLs, and required models. You can zip and share this folder seamlessly.

Quick Start

GUI

cargo build --release
cargo run --release --bin never-miss-gui

The GUI lets you launch Chrome and configure all parameters with a visual interface.

CLI

1. Start Chrome with remote debugging

/Applications/Google\ Chrome.app/Contents/MacOS/Google\ Chrome \
  --remote-debugging-port=9222 \
  --user-data-dir="chrome_profile"

2. Log in and navigate to the event page

Open the event page manually in that Chrome window.

3. Run

# Basic — auto-selects first date, first area, 1 ticket
cargo run --release --bin never-miss -- -w tixcraft

# Full options
cargo run --release --bin never-miss -- -w tixcraft -d "2026/05/15" -a "全區" -n 2 -f

# With pre-sale timing
cargo run --release --bin never-miss -- -w tixcraft -d "2026/05/15" -a "全區" -n 2 -f -t "12:00"

CLI Reference

Flag	Default	Description
-w, --website	interactive prompt	Website profile name from config.yaml
--cdp-url	127.0.0.1:9222	Chrome DevTools Protocol endpoint
-d, --date	auto (first)	Date / session to select
-a, --area	auto (first)	Area or price tier
-n, --tickets	1	Number of tickets
-t, --sale-time	—	Pre-sale time (HH:MM, Taiwan time)
-f, --fast	off	Fast mode — reduced human-like delays
-c, --config	config.yaml	Path to config file

Configuration

config.yaml declares one profile per platform. No credentials are stored.

fast_profile:
  hover_delay: [10, 30]
  click_hold: [15, 40]
  poll_interval: 500

fast_profile:
  hover_delay: [10, 30]       # pause before clicking (ms)
  click_hold: [15, 40]        # mousedown → mouseup duration (ms)
  think_delay: [5, 15]        # pause before cursor starts moving (ms)
  move_delay_fast: [1, 3]     # cursor speed at mid-path (ms per step)
  move_delay_slow: [3, 8]     # cursor speed at start/end (ms per step)
  steps_per_100px: [4, 8]     # waypoints per 100px of travel
  poll_interval: 500          # ms between FSM cycles when stalling

websites:
  - name: tixcraft
    display_name: "TixCraft (拓元售票)"
    flow:
      - state: Preparing
        action: click
        keywords: ["Buy tickets", "立即購票", "購票"]
        url_pattern: "/activity/detail"
      - state: SelectingDate
        action: click
        keywords: ["Find tickets", "立即訂購", "訂購"]
        landmark: "場次"
      - state: SelectingArea
        action: click
        keywords: ["區", "區域", "全區", "area", "樓", "zone"]
        url_pattern: "/ticket/area"
      - state: SelectingCount
        action: select_count
        keywords: ["數量", "Quantity", "張", "＋", "+", "增加", "全票", "票"]
        url_pattern: "/ticket/ticket"
      - state: SolvingCaptcha
        action: captcha_cnn
        keywords: ["verifycode", "checkcode", "驗證碼", "verification code", "請輸入下方驗證碼數字"]
      - state: AcceptingTerms
        action: check_all
        keywords: ["同意", "agree", "accept", "我已閱讀", "閱讀並同意", "確認同意"]
      - state: Submitting
        action: click
        keywords: ["submit", "送出", "確認", "confirm", "下一步", "next"]
      - state: Checkout
        action: wait_url
        keywords: []
        url_pattern: "/ticket/checkout"

For implementation details, see ARCHITECTURE.md.

Demo

The graphical user interface	Demo on YouTube

Disclaimer

This project is for educational and research purposes only. The authors are not responsible for any misuse, account bans, or violations of terms of service resulting from the use of this software. Always comply with the ticketing platform's terms and conditions. Automated ticket purchasing may violate local laws or platform policies — use at your own risk.

License

This project is licensed under the GNU General Public License v3.0 - see the LICENSE file for details.