Your AI-powered PC shopping assistant. Scans your current hardware, benchmarks performance, identifies bottlenecks, and hunts the web for your perfect upgrade — all through natural conversation with Claude.
- Scans your current hardware specs automatically
- Benchmarks your CPU, RAM, and storage performance
- Interviews you about what you need and your budget
- Diagnoses where your current PC falls short (with a letter-grade Report Card)
- Hunts the web for the best deals matching your requirements
- Recommends top picks with full price/performance analysis
- Claude Code (Anthropic's CLI tool)
- Python 3.8+
- Dependencies listed in
requirements.txt
That's it. No extra API keys, no accounts, no complicated setup.
If you don't have Claude Code yet, install it via npm:
npm install -g @anthropic-ai/claude-codeThen authenticate:
claude loginNeed help? See the Claude Code docs.
You need Python 3.8 or newer. Check with:
python --versionIf you don't have it:
- Windows: Download from python.org and check "Add to PATH" during install.
- macOS:
brew install python3(or download from python.org) - Linux:
sudo apt install python3(Ubuntu/Debian) orsudo dnf install python3(Fedora)
git clone <repo-url>
cd Speed-testOr download the ZIP and extract it.
pip install -r requirements.txtThis installs psutil which gives the benchmark script enhanced details like
CPU frequency, available RAM, and disk usage. The core scripts work without it,
but you get richer output with it installed.
Open wishlist.yaml in any text editor and fill in your budget, use case,
and preferences. This is totally optional — if you skip it, Claude will just
ask you these questions during the conversation.
Open a terminal in the project folder and launch Claude Code:
cd Speed-test
claudeThat's it. Claude reads the CLAUDE.md guide automatically and knows exactly
what to do. Just tell it what you need:
- "Scan my PC and tell me what I've got"
- "I need a gaming PC under $1,500"
- "What should I upgrade first?"
- "Find me the best prebuilt with an RTX 5070"
- "Build me a video editing workstation for $2,000"
- "Is my PC good enough for streaming?"
Claude will run the scanner and benchmarks for you, then walk you through the full recommendation process.
You don't have to run these manually (Claude handles it), but you can:
# Scan your hardware — prints a report and saves JSON to results/my_specs.json
python detect.py
# Scan hardware — output raw JSON only
python detect.py --json
# Scan hardware — save JSON silently (no terminal output)
python detect.py --quiet
# Run benchmarks — takes 2-4 minutes, saves to results/benchmark.json
python benchmark.pyHardware scan prints a formatted report like this:
╔════════════════════════════════════════════════════════════╗
║ PC-PILOT HARDWARE SCAN ║
║ Know your machine. Find your upgrade. ║
╚════════════════════════════════════════════════════════════╝
Form Factor: Desktop
──────────────────────────────────────────────────────────────
PROCESSOR (CPU)
──────────────────────────────────────────────────────────────
Model Intel(R) Core(TM) i9-14900KF
Cores / Threads 24C / 32T
Socket LGA 1700
...
Benchmarks score your PC out of 10,000 and assign a tier:
YOUR SCORE: 17,398 / 10,000
TIER: MONSTER
Top 1% -- Extreme enthusiast / workstation
The magic is in CLAUDE.md — a structured guide that teaches Claude Code how
to be a PC hardware consultant. When you open this project in Claude Code, it
follows a 6-phase process:
SCAN --> INTERVIEW --> DIAGNOSE --> HUNT --> COMPARE --> RECOMMEND
| | | | | |
v v v v v v
Detect Ask about Grade each Search Rank by Present
hardware needs & component the web value & top picks
specs budget A+ to F for deals performance with links
Results are saved to the results/ folder for easy reference.
Speed-test/
├── CLAUDE.md # The AI brain — teaches Claude the PC shopping process
├── README.md # You're reading it
├── requirements.txt # Python dependencies (pip install -r requirements.txt)
├── detect.py # Cross-platform hardware scanner (Win/Mac/Linux)
├── benchmark.py # CPU, RAM, and disk performance benchmarks
├── wishlist.yaml # Your preferences template (optional)
├── ram_check.ps1 # Quick RAM usage check (Windows PowerShell)
└── results/ # Output folder (auto-created on first run)
├── my_specs.json # Hardware scan results
├── benchmark.json # Benchmark scores
└── recommendation.md # Final recommendation report
| Platform | Hardware Scan | Benchmarks | Web Search |
|---|---|---|---|
| Windows 10/11 | Yes | Yes | Yes |
| macOS | Yes | Yes | Yes |
| Linux | Yes | Yes | Yes |
What gets detected: CPU (model, cores, threads, clock, socket), GPU (model, VRAM), RAM (total, type, speed, stick config), Storage (model, capacity, NVMe vs SATA vs HDD), Motherboard (manufacturer, model), OS, and form factor (desktop vs laptop).
Known limitations:
- PSU wattage can't be detected via software — check the physical label.
- GPU VRAM for cards >4 GB may use a lookup table if
nvidia-smiisn't available (covers all major NVIDIA/AMD/Intel Arc GPUs). - CPU clock shows base frequency from BIOS, not turbo/boost speed.
- Be specific about your use case. "Gaming" is good. "1440p 144Hz competitive FPS" is better.
- Set a real budget. Include monitor/peripherals if you need those too.
- Mention your timeline. If you can wait, Claude might suggest holding for a sale.
- Ask follow-ups. Claude can deep-dive into any recommendation, compare two specific models, or explain trade-offs.
- Check your PSU. The scanner can't detect PSU wattage. If you're upgrading, check the label on your power supply and tell Claude.
python command not found
Try python3 instead. On Windows, you may need to reinstall Python and check
"Add to PATH."
Unicode errors on Windows (box-drawing characters) The scripts handle this automatically, but if you see garbled output, try running in Windows Terminal (not the old cmd.exe).
nvidia-smi not found
This only matters for NVIDIA GPU VRAM detection. The scanner falls back to a
built-in lookup table for all common GPUs. If your GPU model is missing, VRAM
will show as null — Claude will look it up for you.
Benchmark takes too long Normal runtime is 2-4 minutes. If it hangs, it's likely on the disk benchmark (writes a 512 MB test file). Make sure you have at least 1 GB of free temp space.
Found a bug? Have an idea? Open an issue or PR.
MIT