Local AI Writing Assistant

AI runs on your device. First use downloads open-source model files from Hugging Face. Your input is not sent for inference. The model host receives normal download requests. Model files may be cached by your browser; device speed and memory affect results.

Write with a purpose

TaskText or factual brief

0 / 4,000 characters

Processing mode

Desktop recommended. GPU mode needs WebGPU with shader-f16 support. CPU mode is slower; keep this tab open.

Ready. No model downloads until you start.

Your result

Editable output

Small local models can omit details or invent facts. Check names, numbers and claims against your original material.

Six writing tasks, one local workspace

Choose the task, enter a short English brief, approve the first model download and generate a draft. For product descriptions, supply real features and limitations. The assistant does not research products, verify claims or guarantee plagiarism-free output.

Model and privacy

Powered by Hugging Face SmolLM2-360M-Instruct (Apache 2.0), a compact English model. It is less capable than larger hosted assistants. Text is held in this page, not saved automatically. Stop releases the worker and model memory; browser-cached model files may remain.

Summarize selected PDF text · Transcribe English audio · All AI tools