rasonyang/ai-native-callcenter

What it solves

This project provides an open-source, AI-first call center platform where an AI voice model acts as the primary responder rather than a secondary add-on. It solves the fragmentation of call records by unifying AI-handled interactions and human-transferred calls into a single, continuous conversation history.

How it works

The system uses a Go binary that integrates with FreeSWITCH and PostgreSQL. When a call arrives, it is bridged to a voice model via a speech-to-speech connection. A flow engine manages the conversation phases and tool usage, deciding when to transfer the caller to a human agent. If a transfer occurs, the agent uses a companion Chrome extension to handle the call, while the system maintains a single record containing the full transcript and call data.

Who it’s for

It is designed for organizations that want to build a native AI call center experience, requiring infrastructure knowledge of FreeSWITCH and the ability to manage a Go-based backend. It is suitable for teams needing a unified platform that supports both automated AI dialogue and seamless hand-offs to human agents.

Highlights

• Unified Records: Every call, whether handled by AI or a human, results in one single call detail record with a complete transcript.
• AI-First Dialogue: The AI model owns the conversation flow, using defined phases and tools to steer interactions without relying on rigid scripts.
• Seamless Handoffs: When a human is needed, the call is transferred to a queue, and the agent's browser screen is automatically populated with the call's context and transcript.
• Self-Contained Deployment: The entire application—including the REST API, event stream, and web interface—is packaged as a single Go binary.
• Flexible Provider Support: Supports multiple voice providers (such as OpenAI, Grok, Qwen, or Doubao) and handles both English and Chinese.