Pipecat 1.0 ships Subagents, runs parallel agents with Gradient Bang demo

Pipecat just crossed from a voice agent toolkit into realtime multi-agent infrastructure. Version 1.0 lands after two years. It frames the project as multimodal and multi-model, not just voice. It is already used by thousands of startups, scale-ups, and enterprises, with contributions from NVIDIA, AWS, GCP, Azure, and model labs.

The new piece is Subagents. Multiple inference loops run in parallel and share partial context over an event bus. It works locally and across a network. Instead of just docs, they shipped a live multiplayer game, Gradient Bang, plus full source and prompts. It shows agents coordinating, colliding, and leaking context in real time.