# How PolyAI Uses Kong Gateway to Power Developer-Friendly Conversational AI APIs

This talk explains how PolyAI used Kong to turn a largely UI-driven platform for enterprise customer-service voice agents into a consistent, headless developer platform. The team describes why that shift mattered: customers wanted REST APIs, CLI-based CI/CD workflows, MCP servers, and real-time voice access without rebuilding authentication, authorization, and rate limiting for every new entry point. Their approach was to treat each interface as a new “door” guarded by the same gateway, keeping identity and control centralized while defining APIs declaratively and reusing shared specs across routes, docs, developer portals, and MCP tools. The speakers also outline PolyAI’s voice architecture, contrasting traditional cascaded systems and fully streaming audio models with a hybrid approach that combines audio listening, turn-taking, and reasoning in one step while still returning text for control and lower cost. Key lessons include preserving low latency for natural conversations, especially under 300 milliseconds, and embedding guardrails deeply into specialized customer-experience models.