idle intelligence

llm-web

A chat interface for the lean engine, a Rust language-model engine running client-side via WebAssembly, on WebGPU, CPU threads or a single CPU thread depending on what this browser offers.

github.com/idle-intelligence/llm-web

settings · model

SmolLM2-360M-Instruct Q4_0 (219 MB) Qwen2.5-0.5B-Instruct Q4_0 (430 MB) Qwen3-0.6B Q8_0 (640 MB) Qwen3-1.7B Q8_0 (1.8 GB) SmolLM2-1.7B-Instruct Q4_0 (994 MB)

settings · backend

Auto WebGPU CPU threads Single CPU thread

STATUS

Initializing...

INPUT · message

Everything runs on this device.

OUTPUT · conversation

Replies will appear here...

PERFORMANCE

—

TTFT (ms)

—

DECODE TOK/S

—

PROMPT TOKENS

—

TOKENS