Industry story
Pydantic AI adds realtime speech-to-speech and new cloud providers
agent-framework inference reliability tool-use
Pydantic AI is quietly becoming the connective tissue for production agent infrastructure. Version 0.2.28 adds Agent.realtime() for live speech-to-speech conversations via WebRTC, plus server-side signaling, which moves the framework well past text-only use cases. New provider support for Crusoe and Cerebras means agent builders can now route to GPU-optimized inference backends without leaving the framework. The backend roster keeps growing, and that breadth is what turns a library into a platform.
Full analysis
Alongside the security fix, Pydantic AI v2.28.0 introduces Agent.realtime(), a new API for realtime speech-to-speech interactions, with additional support for browser WebRTC plus server-side signaling. This enables agents to participate in live audio conversations rather than only text-based exchanges. The release also adds a Crusoe provider (a GPU cloud) and a Cerebras optional dependency group, expanding the range of inference backends agent builders can target directly from the framework.
Comments