Running an LLM in Your Browser: Verifying WebGPU, Model Hashes, and Local Inference Without a Server

WebGPU reached stable cross-browser support in 2025, enabling LLMs from Hugging Face's Transformers.js v4 to run fully in-browser. Open-weights models (Llama, Phi, Qwen) in the 1B–8B range with 4-bit quantization occupy 0.8–5 GB, fitting consumer hardware. CapyToolkit provides free browser tools for verification: a hash verifier for SHA-256 checksums, a token counter for tokenizer cross-checks, and a fingerprint inspector. Users should also monitor DevTools' Network tab for outbound leaks.