🎤 VoiceCLI

VoiceCLI is a command-line tool that listens to your voice, transcribes it, uses a local Ollama LLM to convert the transcription into a safe shell command, and then executes it after confirmation.

---

🧠 How It Works

1. Records 5 seconds of audio.
2. Transcribes it using whisper.cpp.
3. Sends the transcription to a local Ollama model (like llama3).
4. Returns a shell command (or # unknown).
5. Asks for your confirmation.
6. Executes the command.

---

🛠️ Requirements

• Go 1.20+
• sox (macOS) or arecord (Linux)
• whisper.cpp built as main
• Whisper model (e.g. ggml-base.en.bin) in models/
• Ollama running locally with a pulled model:

  ollama pull llama3

---

🚀 Run It

go run ./cmd/voicecli

---

📁 Project Structure

voicecli/
├── cmd/voicecli.go         # CLI entry point
├── internal/
│   ├── audio/              # Records audio from mic
│   │   └── recorder.go
│   ├── whisper/            # Transcribes speech
│   │   └── transcriber.go
│   ├── ollama/             # Sends prompt to local Ollama LLM
│   │   └── client.go
│   └── shell/              # Confirms and executes commands
│       └── executor.go

---

🛡️ Safety

• Uses a strict system prompt with Ollama to only return simple bash commands.
• Asks for user confirmation before running anything.

---

🧪 Example

🎤 VoiceCLI is listening... (press Ctrl+C to exit)
🗣️  You said: list my files
💻 Generated command: ls -la
⚠️  Do you want to execute this command? [y/N]: y
✅ Done.

---

📌 TODO

• Continuous listening mode
• Better error recovery and retries
• Add Windows audio support (e.g., WASAPI)
• Customizable system prompt

---

💡 Author

Abdulrahman Masoud, [@AbdulrahmanMasoud](

Made with ❤️ using Go, Whisper, and Ollama.