BusDK Update
Offload dev workers to a remote GPU box over SSH
BusDK's dev-worker stack gained an early remote environment path for running agent work on a GPU machine over SSH. The pieces are deliberately practical: remote worker environments, rootless Docker access on the remote GPU host, and a Codex app-server proxy for the worker connection.
For dedicated, customer-controlled deployments, this lets the local BusDK stack drive work while execution runs on hardware the operator controls. The shape is experimental, so it fits controlled trials where the operator owns the host, the SSH boundary, and the Docker setup.
May 25, 2026remote workersSSH offloadrootless Docker
The deployment shape keeps the BusDK control plane local and moves worker execution to a remote machine. The remote environment supplies a checkout, container runtime access, and a socket path usable by worker containers. SSH is the operational boundary for setup and synchronization, so the GPU box can stay inside the operator's own account and network policy.
ssh gpu-box.example 'test -S /run/user/1000/docker.sock && echo rootless-docker-ready'
rootless-docker-ready
The Codex app-server proxy is the connection piece. It gives the worker process a stable app-server endpoint even when the execution host is remote. That keeps task submission and observation in the local stack while letting the worker substrate run on a dedicated remote GPU machine.
This maps to BusDK's dedicated, customer-controlled environment deployment tier. A team can bring its own remote GPU box, run the worker substrate there, and keep the local BusDK workflow as the driver. The maturity is early: expect environment-specific setup, SSH and Docker assumptions, and careful operator validation before routine production use.
See docs.busdk.com for BusDK module documentation.