reaatech/voice-agent-kit

These packages provide a transport layer for real-time voice AI, orchestrating the pipeline between telephony streams, speech-to-text, text-to-speech, and MCP-based agent logic. They solve the challenge of maintaining sub-second latency in conversational systems by enforcing strict timing budgets and providing provider-agnostic adapters for services like Deepgram, AWS, and Google Cloud. The architecture centers on an event-driven pipeline that decouples telephony handling from agent decision-making, allowing you to swap providers or integrate custom MCP servers without modifying core orchestration logic.