AI Hardware Check

A browser-only tool that detects your system hardware and estimates which open-source AI models you can run locally — no uploads, no accounts, no backend.

Features

• Model library — curated catalog of 40+ model families (Meta, Mistral, DeepSeek, Qwen, Google, and more), grouped by provider and family
• Guided setup assistant — choose your goal (chat, coding, vision, offline) and deployment style (local app, CLI, hosted API) to get a plain-language next step
• Add any model — paste a HuggingFace repo ID and the app fetches parameter count, modality, and provider automatically, then groups it alongside the built-in catalog
• RAM fit estimation with reasoning — compares your RAM against quant requirements and explains why each quant is a good, borderline, or bad fit
• Confidence-scored hardware profiling — reports not only specs, but also source quality (reported vs measured), unresolved gaps, and an inferred performance tier
• Performance range estimation — predicts a conservative-to-expected token/s range plus first-token latency instead of a single optimistic value
• Quantization table — every quant row links directly to the HuggingFace repo or a pre-filled search for community GGUF downloads
• System hardware snapshot — CPU threads, RAM, GPU renderer, WebGL / WebGPU status, all from browser APIs
• Search & filters — filter by provider, compatibility status, or modality; full-text search across model names, families, and repo IDs
• Dark / light theme — persisted to localStorage
• 100 % client-side — no server, no data collection

Quick Start (Docker Hub)

# Pull and run in one step
docker run --rm -p 38421:80 gptvibe/ai-hardware-check

Then open http://localhost:38421.

docker-compose

# Clone the repo and run with compose
git clone https://github.com/gptvibe/AI-Hardware-Check.git
cd AI-Hardware-Check
docker compose up -d

By default, compose exposes the app on port 38421. To use a different port:

HOST_PORT=40123 docker compose up -d

Local Development

git clone https://github.com/gptvibe/AI-Hardware-Check.git
cd AI-Hardware-Check
npm install
npm run dev        # http://localhost:5173

Production Build

npm run build      # output in dist/
npm run preview    # serve the built output locally

Docker (build from source)

docker build -t ai-hardware-check .
docker run --rm -p 38421:80 ai-hardware-check

Use any host port you want by changing the left side of -p, for example:

docker run --rm -p 40123:80 ai-hardware-check

Adding Models

Click the input bar at the top of the page and paste any public HuggingFace repo ID, for example:

Qwen/QwQ-32B
mistralai/Mistral-Small-3.2-24B-Instruct-2506
google/gemma-3-27b-it

The app fetches the safetensors parameter count and pipeline tag, derives the provider from the org prefix, estimates RAM requirements, and persists the entry in localStorage across reloads. Custom models can be removed from the detail panel.

Model Catalog

Models are stored in public/models.json. Each entry supports:

Field	Description
name	Display name
provider	Company / org (used for grouping)
family	Model family name (optional, defaults to name)
parameter_count	e.g. 8B, 70B, 1T
huggingface_repo	org/model-name
modalities	["Text"], ["Image","Text"], etc.
formats	Quantization formats available
ram_requirements_gb	Optional overrides per quant key
active_params_b	Active parameters in billions for MoE-aware speed estimates (optional)
context_windows	Supported context lengths, e.g. [4096, 8192, 32768] (optional)
runtime_recipe_templates	Optional command templates for ollama, lmstudio, llamacpp
release_date	ISO date string for recency sorting (optional)
quant_download_links	Explicit download URLs per quant key (optional)
notes	Short note shown in the UI (optional)

RAM is auto-estimated from parameter_count when ram_requirements_gb is not provided.

The app also supports an optional 20-second local calibration benchmark to improve token/s prediction ranges for your specific machine.

License

MIT