Claude
cc847cff67
When run via `bash <(curl ...)`, bash reads the script incrementally from a process substitution fd. If a `read < /dev/tty` call disrupts the fd state, bash can lose its place and the remaining script lines get fed to the interactive shell prompt as individual commands. Wrapping everything in main() forces bash to parse the entire script into memory before executing any of it — the standard fix for curl-piped scripts. Also changed README one-liner to download-then-execute pattern (curl -o /tmp/... && bash /tmp/...) as a belt-and-suspenders approach. https://claude.ai/code/session_01XKYC1basxdwtHy71tky7xm
4.2 KiB
4.2 KiB
OpenWhispr Easy Setup
One-command installer for OpenWhispr on Linux — distro detection, dependencies, API keys, first-run setup.
Your voice stays on your machine. Always. The only question is whether you want AI to clean up the text after transcription.
Quick install
curl -fsSL https://raw.githubusercontent.com/outis1one/openwhispr-easy-setup/main/setup-openwhispr.sh -o /tmp/setup-openwhispr.sh && bash /tmp/setup-openwhispr.sh
Or clone and run:
git clone https://github.com/outis1one/openwhispr-easy-setup.git
cd openwhispr-easy-setup
bash setup-openwhispr.sh
What it does
| Step | Action | Details |
|---|---|---|
| 1/8 | Intro | Explains OpenWhispr and the privacy model |
| 2/8 | Detect distro + arch | Ubuntu, Debian, Mint, Fedora, openSUSE, Arch — amd64/arm64 |
| 3/8 | Install OpenWhispr | Fetches latest release from GitHub, picks correct package |
| 4/8 | Paste dependencies | xdotool (X11), wtype (Wayland), kdotool (KDE), or D-Bus (GNOME) |
| 5/8 | STT provider | Local Whisper/Parakeet, Groq, or OpenAI Realtime |
| 6/8 | AI cleanup | OpenAI, Anthropic, Groq, Ollama, or skip |
| 7/8 | First launch | Opens OpenWhispr for model download and hotkey setup |
| 8/8 | Summary | Prints config, useful commands, and next steps |
STT options
| Option | Method | Streaming | Cost | Audio sent to |
|---|---|---|---|---|
| 1 (default) | Local Whisper/Parakeet | No | Free | Nobody, ever |
| 2 | Groq whisper-large-v3-turbo | No | $0.04/hr* | Groq servers |
| 3 | OpenAI Realtime API | Yes | $0.06/min | OpenAI servers |
*Groq free tier: ~2,000 audio seconds/day.
AI cleanup options
| Provider | Model | Per use | 50/day/mo | 200/day/mo |
|---|---|---|---|---|
| OpenAI | gpt-4o-mini-2024-07-18 | $0.000058 | $0.09 | $0.35 |
| Anthropic | claude-haiku-4-5 | $0.000450 | $0.68 | $2.70 |
| Groq | llama-3.3-70b-versatile | Free* | Free* | Free* |
| Ollama | Local (fully offline) | Free | Free | Free |
| Skip | Raw transcription only | Free | Free | Free |
*Groq free: 1,000 req/day, 100K tokens/day.
Privacy
| Data | Where it goes | Condition |
|---|---|---|
| Audio | Stays on your machine | Default (local Whisper/Parakeet) |
| Audio | Groq or OpenAI servers | Only if you choose cloud STT |
| Text | AI cleanup provider | Only if you choose AI cleanup |
| API keys | ~/.bashrc on your machine |
Never sent anywhere by this script |
No provider uses your data for training. Cloud data is deleted within 30 days.
Useful commands
openwhispr # Launch the app
openwhispr --help # Show options
source ~/.bashrc # Reload API keys in current shell
In the app:
- Settings → Hotkey — change your hotkey (default: backtick
`) - Settings → Models — download better speech models
- Settings → Processing — change STT or AI provider
Supported distros
| Distro | Package format | Status |
|---|---|---|
| Ubuntu / Debian / Mint | .deb |
Supported |
| Fedora | .rpm (dnf) |
Supported |
| openSUSE | .rpm (zypper) |
Supported |
| Arch / Manjaro / EndeavourOS | .tar.gz |
Supported |
Architecture: x86_64 (amd64) and aarch64 (arm64).
Requirements
- bash 4+
- curl — for downloading
- jq — for parsing GitHub API (script offers to install if missing)
Safe to re-run
The script is idempotent:
- Skips OpenWhispr install if the latest version is already present
- Never overwrites existing API keys in
~/.bashrc - Detects already-installed paste dependencies