voxai 0.2.4

Voice-driven AI assistant for real-time transcription and Gemini integration.

VoxAI

🚀 Features

• Universal Audio Capture
  Record mic, system audio, or any combination via BlackHole (macOS) or equivalent loopback drivers.
• Manual Control
  Start/Stop buttons let you define exactly the boundaries of your prompt—perfect for long, multi-sentence queries.
• One-Shot Transcription
  Whisper processes the entire recording in a single call—no fragmented sentences.
• Live AI Streaming
  Gemini’s response appears token by token, just like in ChatGPT’s streaming interface.
• Configurable Model
  Swap between Gemini variants via environment (no code changes).
• Zero-Install UI Bootstrapping
  On first run voxai auto-installs the minimal Electron UI source—included in the PyPI package—so you only ever need pip install voxai and voxai.

---

🎯 Quickstart

⚙️ Prerequisites

• Python ≥3.7

• Node.js ≥14

• Electron (global)

  npm install -g electron
  

• macOS Audio Loopback (BlackHole 2ch + Ladiocast):

  1. Install BlackHole 2ch::

     brew install blackhole-2ch

  2. Create a Multi-Output Device

     • Open Audio MIDI Setup
     • Click “+” → Create Multi-Output Device
     • Check both BlackHole 2ch and MacBook Pro Speakers

  3. Select Multi-Output Device
     System Preferences → Sound → Output → Multi-Output Device

  4. Configure Ladiocast

     • Input 1: BlackHole 2ch → route to Main
     • Main Output: MacBook Pro Speakers
     • Mute other channels

  This will route all system and microphone audio into VoxAI.

Install & Run

1) Install from PyPI

pip install voxai

2) Configure environment

cp .env.example .env

Edit .env:

GENAI_API_KEY=sk-…

GENAI_MODEL=gemini-1.5-flash

3) Launch the app

voxai

On first launch, VoxAI will automatically run npm install inside its bundled electron/ folder, then open a desktop window.

---

📖 Usage

Once the UI opens:

Start: Click Start to begin recording.

Speak: Listen to question`s from zoom, teams or any other tool aloud—no length limit.

Stop: Click Stop. Whisper transcribes your entire clip to the Transcript pane.

Ask AI: Click Ask AI. Gemini’s answer streams live into the Answer pane.

Copy & Share: Everything is plain text—copy it, paste it, or feed it into your own RAG/finetuning pipeline.

---

🎧 How It Works

• Start Recording
  Click Start to capture audio from your input device (e.g., BlackHole).

• Stop & Transcribe
  Click Stop. Whisper transcribes and shows the full audio clip under Transcript.

• Ask AI
  Click Ask AI to send the transcript to Gemini. Answers stream live in the Answer panel.

• Copy & Share
  Output is plain text—reuse it in RAG/finetune workflows or anywhere else.

---

🛠 Configuration

Set in .env:

Variable	Description	Example
GENAI_API_KEY	Google Generative AI (Gemini) API key	sk-…
GENAI_MODEL	Gemini model to use	gemini-1.5-flash

See .env.example for reference.

---

🛠️ Developer Guide

Install from source

git clone https://github.com/rtiwariops/voxai.git

cd voxai

pip install -e .

npm install -g electron

voxai

⚙️ Features

• Universal Audio Capture (BlackHole, etc.)
• Manual Control for long-form Q&A
• One-Shot, Full-Clip Transcription
• Live Token-by-Token AI Streaming
• Configurable Gemini Model (.env)
• Electron-Based Desktop UI

---

📜 License

MIT License © 2025 Ravi (Robbie) Tiwari

Released under the MIT License.