Gander Splits a Full-Duplex Omni Agent Into a Realtime Cerebellum and a Reasoning Brain
Gander is an end-to-end model unifying omni perception, realtime interaction and agentic capability, continuously receiving streaming video, speech and text so users can interrupt at any time while the model proactively offers intermediate feedback or asks follow-up questions. It uses a Cerebellum-Brain collaborative framework where the Cerebellum handles realtime interaction and conversation while the Brain handles complex reasoning and agentic tasks, the two communicating through tool calling and an agent orchestration runtime. The Cerebellum runs a streaming Thinker-Talker architecture flattening user inputs and model outputs into an ordered chunk-level token stream, and the models, code and data are released; the paper drew 88 HuggingFace upvotes on its first day.
↳ Follow the thread