VVoxGraph

Quick start

Get a full voice loop running on your machine: speak → STT → LLM → TTS → hear the reply.

Prerequisites

  • Python 3.10+
  • Deepgram API key (STT + Aura TTS)
  • Ollama (local LLM, recommended) or Google Gemini API key
  • Headphones (reduces echo from speaker → mic)

1. Clone and install

bash
git clone https://github.com/suryaraj05/voxgraph.git
cd voxgraph
python -m venv .venv
.venv\Scripts\activate   # Windows
# source .venv/bin/activate  # macOS/Linux
pip install -r requirements.txt

2. Configure environment

bash
cp .env.example .env

Edit .env. Minimum for local demo:

env
DEEPGRAM_API_KEY=your_key
LLM_PROVIDER=ollama
OLLAMA_MODEL=qwen2.5:7b
TTS_PROVIDER=deepgram
DEMO_MODE=1
STT_ONLY=0

3. Start Ollama (if using local LLM)

bash
ollama serve
ollama pull qwen2.5:7b

4. Run the server

bash
python -u -m uvicorn voxgraph:app --host 0.0.0.0 --port 8001

5. Run the live mic client

In a second terminal, from the voxgraph folder:

bash
python scripts/live_mic_client.py --demo --url ws://127.0.0.1:8001/audio

Speak naturally. The server detects end-of-utterance, calls the LLM, and streams TTS back. Type q + Enter to quit.

Verify it works

  • Server log: Deepgram connection opened
  • Client log: Live playback on (24000 Hz PCM)...
  • After you speak: · Thinking… → · Speaking… on the client

Next steps