convonetai
voice AI for mortgage application search
Core trick
Wraps Cartesia batch STT and TTS around a Claude planner, executor, and critic loop with SQLite-backed memory.
Description
voice AI for mortgage application search
README excerpt
# Cartesia Voice Agent (Convonet-Style Scaffold) Open-source-friendly scaffold for a **Cartesia STT/TTS + Anthropic Claude + Notion MCP** voice agent that follows the **Convonet-style** planner → executor → critic loop. This repo is designed for hackathon iteration: minimal but functional, with clear extension points. ## What This Implements - **Cartesia STT (Batch)** using `https://api.cartesia.ai/stt` (no Whisper). - **Cartesia TTS** using the official Python SDK. - **Claude** tool-calling loop (Anthropic) for planning + execution. - **Notion MCP + Custom Agent** tool integration (optional). - **Memory store** (SQLite) for short-term + long-term context. ## Quick Start 1. Create a virtual environment and install deps: - `python -m venv .venv && source .venv/bin/activate` - `pip install -e .` 2. Copy env template: - `cp .env.example .env` 3. Fill in API keys and Notion MCP settings. 4. Run a basic turn: - `voice-agent --audio ./samples/input.wav --output ./samples/response.wav` - or `voice-agent --text "Search and find the best mortgage program based on my credit score and income" --output ./samples/response.wav` ## Demo Server (WebRTC/Voice Assistant UI) Start the demo server and open the WebRTC UI entry point: - `voice-agent-server` - Open: `http://localhost:8000/webrtc/voice-assistant` This serves the Convonet call center UI and exposes stubbed `/call-center/api/*` endpoints, plus a demo voice agent endpoint at `/call-center/api/assistant/respond`. #