BYOB (Bring Your Own Brain) CLI tool for audio/video knowledge extraction. Extracts speech from media, assembles expert prompts, and delivers them to your clipboard — ready for any LLM.
YouTube URL / local file
│
▼
yt-dlp + ffmpeg (audio extraction)
│
▼
ASR engine (Groq Whisper API or sherpa-onnx local)
│
▼
Prompt assembly (language-aware, tail-placement)
│
▼
Clipboard / git archive
# via cargo-binstall (prebuilt binary)
cargo binstall lpdwise
# or build from source
cargo install --path crates/lpdwise# Check that dependencies are available
lpdwise doctor
# Extract knowledge from a YouTube video
lpdwise https://www.youtube.com/watch?v=...
# Extract from a local audio file
lpdwise /path/to/recording.mp3Configuration file: ~/.config/libre-pdwise/config.toml
groq_api_key = "gsk_..."
# data_dir = "/custom/path" # optional override| Variable | Description | Priority |
|---|---|---|
GROQ_API_KEY |
Groq Whisper API key | Highest (overrides config file) |
LPDWISE_DATA_DIR |
Data directory override | Highest |
HF_MIRROR |
HuggingFace mirror URL for model downloads | - |
~/.local/share/libre-pdwise/
├── media/ # downloaded audio
├── archive/ # git-based transcript archive
├── models/ # sherpa-onnx model files
└── logs/ # structured logs
| Engine | Type | Best For | Requires |
|---|---|---|---|
| Groq Whisper | Cloud API | English, fast turnaround | GROQ_API_KEY |
| sherpa-onnx SenseVoice | Local | Chinese speech | 2GB+ RAM |
| sherpa-onnx Whisper | Local | General fallback | 2GB+ RAM |
Engine selection is automatic based on detected language and available resources.
| Template | Purpose |
|---|---|
| Standard | Structured summary with outline, key points, and source quotes |
| Contrarian | Extract counterintuitive claims with evidence assessment |
| Political | Political-economic logic decomposition: actors, interests, game theory |
| Translation | Full faithful translation into Chinese with terminology annotations |
- Cloud ASR (Groq): audio is sent to Groq's API for transcription. See Groq's privacy policy.
- Local ASR (sherpa-onnx): all processing happens on-device. No data leaves your machine.
- Archive: transcripts are saved locally via git. Opt out with
--no-archive. - No telemetry: libre-pdwise does not collect or transmit usage data.
MIT