Voice-to-text dictation app with local (Nvidia Parakeet/Whisper) and cloud models (BYOK). Privacy-first and available cross-platform.
JavaScript aianthropiccross-platformgeminigroqlinuxmacosnvidiaopen-sourceopenaiparakeetspeech-to-texttranscribewhisperwindows
docs: GPU acceleration notes, in-app meeting prompts, URL-import network hosts
Gabriel Stein51a19ed29da7
diff --git a/LOCAL_WHISPER_SETUP.md b/LOCAL_WHISPER_SETUP.mdindex d930cbd65..6545a0fef 100644--- a/LOCAL_WHISPER_SETUP.md+++ b/LOCAL_WHISPER_SETUP.md@@ -21,6 +21,17 @@ The first transcription will download the model automatically.| small | 466MB | Medium | Better | ~2GB | Professional use || medium | 1.5GB | Slow | High | ~5GB | High accuracy || large | 3GB | Slowest | Best | ~10GB | Maximum quality |+| turbo | 1.6GB | Fast | High | ~6GB | Fast + accurate |++## GPU Acceleration++Local Whisper can run on your GPU for much faster transcription:++- **macOS**: Metal acceleration is built in — no setup needed on Apple Silicon+- **NVIDIA (Windows/Linux)**: one-click CUDA runtime download from the GPU card in the transcription model picker+- **AMD / Intel (Windows/Linux)**: one-click Vulkan runtime download from the same GPU card — covers Radeon and Arc/integrated GPUs++The GPU runtime is downloaded on demand with SHA-256-verified checksums. If the GPU server crashes or fails to start (unsupported GPU, out of VRAM), OpenWhispr automatically falls back to CPU transcription and shows a notice — dictation keeps working.## How It Worksdiff --git a/TROUBLESHOOTING.md b/TROUBLESHOOTING.mdindex 6d900ecea..b45e8eb90 100644--- a/TROUBLESHOOTING.md+++ b/TROUBLESHOOTING.md@@ -102,6 +102,8 @@4. Clear model cache: `rm -rf ~/.cache/openwhispr/whisper-models`5. Try cloud transcription as fallback+**GPU acceleration (CUDA / Vulkan):** If the GPU-accelerated whisper-server crashes at startup (unsupported GPU, out of VRAM), OpenWhispr automatically restarts it on CPU, retries the same request, and shows a "using CPU instead" notice — the dictation still completes. GPU acceleration can be toggled off from the GPU card in the transcription model picker.+### Wayland Clipboard Issues (Linux)**Symptoms:** Paste simulation succeeds but target app shows "clipboard is empty", "no image on clipboard", or "contents not available in the requested format"@@ -155,7 +157,7 @@ OpenWhispr tries clipboard methods in order: `wl-copy` (most reliable) → rende1. Check that meeting detection is enabled in settings2. Verify your OpenAI API key is valid (required for Realtime API transcription)3. Ensure your meeting app (Zoom, Teams, FaceTime) is running — process detection looks for known meeting applications-4. If auto-detection fails, you can manually start recording from the meeting notification+4. If auto-detection fails, you can manually start recording from the in-app meeting prompt (an always-on-top overlay card — it works even with Focus/Do Not Disturb on and never appears in screen shares)### Agent Mode Issuesdiff --git a/docs/network-allowlist.md b/docs/network-allowlist.mdindex 8e4d5b6e0..9f86d83b0 100644--- a/docs/network-allowlist.md+++ b/docs/network-allowlist.md@@ -49,6 +49,18 @@ Contacted only if the user connects Google Calendar in settings.| `www.googleapis.com` | HTTPS | 443 | Calendar event and calendar list reads. || `openwhispr.com` | HTTPS | 443 | OAuth desktop callback redirect (`/auth/desktop-callback`). |+## Required for URL audio import (optional feature)++Contacted only when a user pastes a URL into the Upload view to download and+transcribe its audio. Downloads are HTTPS-only and hosts resolving to+private/internal addresses are rejected.++| Host | Protocol | Port | Purpose |+| ----------------------------------- | -------- | ---- | -------------------------------------------------------------------------- |+| `www.youtube.com`, `youtube.com`, `youtu.be`, `m.youtube.com`, `music.youtube.com` | HTTPS | 443 | YouTube page/metadata fetch for pasted YouTube links (bundled yt-dlp). |+| `*.googlevideo.com` | HTTPS | 443 | YouTube media CDN — the actual audio stream download. |+| _User-pasted hosts_ | HTTPS | 443 | Direct audio/video URL imports contact whatever public host the user pastes. |+## BYOK provider hosts (only if configured)Required only when a user configures their own API key for the corresponding