https://github.com/rollecode/english-refiner
A fast TUI tool for refining English text using LLM via OpenRouter. Fixes grammar while preserving your voice.
https://github.com/rollecode/english-refiner
Last synced: 27 days ago
JSON representation
A fast TUI tool for refining English text using LLM via OpenRouter. Fixes grammar while preserving your voice.
- Host: GitHub
- URL: https://github.com/rollecode/english-refiner
- Owner: rollecode
- Created: 2026-03-24T15:43:41.000Z (2 months ago)
- Default Branch: master
- Last Pushed: 2026-04-13T15:26:08.000Z (about 2 months ago)
- Last Synced: 2026-05-02T13:33:29.202Z (28 days ago)
- Language: Shell
- Size: 14.6 KB
- Stars: 1
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- Changelog: CHANGELOG.md
Awesome Lists containing this project
README
# English Refiner CLI
**A fast TUI tool for refining English text using LLM via OpenRouter. Fixes grammar while preserving your original voice and meaning.**
## Dependencies
- [gum](https://github.com/charmbracelet/gum) - TUI components
- [jq](https://jqlang.github.io/jq/) - JSON parsing
- curl
- [OpenRouter](https://openrouter.ai/) API key
## Installation on macOS
```bash
brew install gum jq
mkdir -p ~/.config/englishrefiner
echo -n 'your-openrouter-api-key-here' > ~/.config/englishrefiner/key
chmod 600 ~/.config/englishrefiner/key
sudo cp englishrefiner /usr/local/bin/englishrefiner
sudo chmod +x /usr/local/bin/englishrefiner
```
## Installation on Linux
Install dependencies for your distro. For Wayland, install `wl-clipboard`. For X11, install `xclip` instead.
```bash
# Arch
sudo pacman -S gum jq wl-clipboard
# Debian/Ubuntu
sudo apt install gum jq wl-clipboard
```
```bash
mkdir -p ~/.config/englishrefiner
echo -n 'your-openrouter-api-key-here' > ~/.config/englishrefiner/key
chmod 600 ~/.config/englishrefiner/key
sudo cp englishrefiner-linux /usr/local/bin/englishrefiner
sudo chmod +x /usr/local/bin/englishrefiner
```
## Usage
```bash
# TUI mode - opens editor, type/paste text, Ctrl+D to submit
englishrefiner
# Clipboard mode - refines clipboard contents, copies result back
englishrefiner -c
# Pipe mode
echo 'your text' | englishrefiner
```
Result is always auto-copied to clipboard.
## Model
Currently using `openai/gpt-5-chat` via OpenRouter. Change the `MODEL` variable in the script to swap models.
## Model benchmarks
Tested with a message containing typos, comma issues, and wrong article usage. Quality rated on whether meaning is preserved while improving readability.
| Model | Speed | Quality | Notes |
|-------|-------|---------|-------|
| google/gemini-2.0-flash-lite-001 | 0.8s | Perfect | Fast, consistent, preserved meaning |
| anthropic/claude-3-haiku:beta | 1.1s | Good | Occasionally adds preamble |
| openai/gpt-5.4-nano | 0.8s | Good | Inconsistent article/dash usage |
| inception/mercury-2 | 0.7s | Good | Inconsistent speed and wording |
| anthropic/claude-3.5-haiku:beta | 2.2s | Good | Same as claude-3-haiku, slower and pricier |
| openai/gpt-oss-120b | 3.5s | Good | Clean output, too slow |
| google/gemini-2.5-flash-lite | 0.7s | OK | Sometimes changes word choices |
| z-ai/glm-5-turbo | 9.5s | OK | Too slow |
| bytedance-seed/seed-1.6-flash | 3.6s | OK | Emdashes, inconsistent |
| stepfun/step-3.5-flash | 5.0s | OK | Too slow, emdashes |
| xiaomi/mimo-v2-flash | 1.1s | OK | Wildly inconsistent speed (1-12s) |
| openai/gpt-4.1-nano | 0.8s | Poor | Changed meaning |
| liquid/lfm-2-24b-a2b | 1.0s | Poor | Rewrote sentences, changed meaning entirely |
| mistralai/ministral-3b-2512 | 0.4s | Poor | Fastest but rewrites everything |
| google/gemini-3.1-flash-lite-preview | 6.1s | Poor | Expanded contractions, too formal |
| openai/gpt-5-nano | 14.8s | Poor | Way too slow |
| qwen/qwen3.5-122b-a10b | 52.7s | Poor | Too slow |
| qwen/qwen3.5-35b-a3b | 55.5s | Poor | Too slow |
| qwen/qwen3.5-flash-02-23 | 77.0s | Poor | Too slow |
| qwen/qwen3.5-9b | 103.6s | Poor | Too slow |
| bytedance-seed/seed-2.0-mini | 122.9s | Poor | Too slow |
| bytedance-seed/seed-2.0-lite | 55.4s | Poor | Too slow |
| z-ai/glm-4.7-flash | 28.1s | Poor | Too slow |