The open-source framework for serving robot policies from remote GPUs over wide-area networks. An inference layer (CUDA graphs with the same outputs, batching within each robot's slack, replicas), a network layer (a robot's connection to its GPU), and the client for GPUs hosted by General Sequences. Supports all open-source robot policies.
python sdk robotics latency websocket scheduling inference pytorch batching vla robot-learning remote-inference openpi vision-language-action lerobot gr00t action-chunking cuda-graphs policy-serving
-
Updated
Nov 19, 2026 - Python