ProofAI

An NVDA add-on that proofreads the text you have selected in any edit field using an OpenAI-compatible LLM endpoint, and replaces the selection with the corrected text.

A small tool built quickly to scratch a real itch — fixing typos and grammar in place, without leaving the field you are writing in — that turned out genuinely useful in daily use.

Install

1. Download the latest .nvda-addon file from Releases.
2. Open it (Enter) to install, and restart NVDA.

Setup

1. NVDA menu → Preferences → Settings → ProofAI.
2. Fill in the API key, endpoint URL, and model.

Defaults: endpoint https://api.openai.com/v1/chat/completions, model gpt-4.1-mini. Any OpenAI-compatible service works — OpenRouter, or a local server such as LM Studio or Ollama.

Usage

1. Focus an edit field and select the text to proofread.
2. Press NVDA+Shift+X.
3. The selection is replaced with the improved text.

The gesture can be changed under Preferences → Input gestures.

Privacy

Only the selected text is sent, and only to the endpoint you configured. Nothing else leaves your machine.

Building from source

python build_addon.py

This produces the installable .nvda-addon package.

License

GPL-2.0, the standard license for NVDA add-ons.