This private AI writer drafts and polishes text without sending it anywhere. Download a small AI model once to write emails, rewrite or shorten text, change tone, summarise, brainstorm or write social posts and product descriptions, all on your device; use Chrome's built-in AI where available; or use the no-download templates, clean-up and summariser. Process a folder of text files at once.
| Engine | Download | Best for |
|---|---|---|
| Qwen2.5 0.5B | ≈ 500 MB once | Emails, rewrites, many languages |
| SmolLM2 360M | ≈ 270 MB once | Quick English drafts |
| Built-in | None | Templates, clean-up, summaries |
No. The model runs in your browser; nothing is sent to a server.
Good at short emails, rewrites and summaries; weaker at facts and long documents, so always read the result.
A recent laptop works; a GPU with WebGPU makes it much faster.
No. Files and text are processed in your browser. Only the model itself is downloaded, once, when you ask for it.
Yes. The Downloaded models panel lists what this tool stored and deletes it after a warning; you'd need to download it again to use it.
Three engines are offered. A small instruction-tuned language model (Qwen2.5-0.5B-Instruct or SmolLM2-360M-Instruct, 4-bit) runs as a Transformers.js text-generation pipeline in a Web Worker on WebGPU or CPU, with greedy decoding, a repetition penalty and tokens streamed to the page. Where the browser provides the Prompt API (Chrome's Gemini Nano), it is used instead. The built-in engine needs no model: fill-in templates for common letters, a deterministic clean-up of spacing, capitals and doubled words, and an extractive summary that picks the highest-scoring sentences by word frequency. Text files can be processed in a batch.