March 26, 2025
It's just that the realtime stuff is so new. Everybody is trying to design the right APIs and abstractions. Experimentation/iteration is good!
You can, of course, use @pipecat_ai to have a common interface to both of those APIs!
➡️ Server-side orchestration with Pipecat core
➡️ All the Pipecat client-side SDKs (js, React, iOS, Android, C++) support direct connections to the Multimodal Live API and OpenAI Realtime WebRTC API.
Can someone explain why @OpenAI's Realtime and @GoogleAI's Gemini Live do not have the same websocket architecture? Why is it so hard to have consistency?!!??!?!