Tether
A deployment tool for vision-language-action robot policies that verifies a model exports correctly, proves it matches the reference implementation, then promotes it to hardware profiles such as Jetson, RTX or Apple Silicon.
Tether is deployment infrastructure, from FastCrest, for vision-language-action models used in robotics. It is built around one production question: whether a given policy can safely move forward to deployment. Its small public workflow chats with the tool, proves an exported model against evidence gathered from the export, then promotes the proof to a target hardware profile such as a warehouse-safe configuration.
The project reports verified numerical parity between its ONNX export and the reference PyTorch policy across four open vision-language-action models: SmolVLA, pi0, pi0.5 and GR00T N1.6, matching to machine precision on shared seeded inputs. It targets Jetson, RTX, Apple Silicon and AMD hardware, with a bootstrap installer that detects the platform and installs the matching extras, or a set of manual installs for GPU or CPU-only serving.
It requires Python 3.10 or newer.