I run a Beam orchestrator and worker on Bittensor Subnet 105. This section records the setup and checks I keep close when maintaining them. BeamCore assigns transfer work, the orchestrator passes it to connected workers, and workers move the data between storage endpoints.
What changed in the current runtime
- The orchestrator and worker are now Go binaries. The old Python launch commands in my earlier guides are outdated.
- BeamCore talks to orchestrators over NATS. Workers connect to their orchestrator over BeamLink/WCP with TLS.
- Subnet registration, BeamCore registration, and worker membership are separate steps.
- The Go processes expose owner-local
/healthzendpoints. A healthy process still needs a working NATS or WCP session before it can receive work.
The path I check is BeamCore → NATS → orchestrator → BeamLink/WCP → worker. The worker then reads from the source and writes to the destination. Local health, network connection, and completed work are separate checks.
My Beam guides
⚡ 🖥️
Orchestrator
Build, register, connect workers, and check readiness.
Worker
Register a worker, join a pool, and check its WCP session.
For API and storage-provider details, use the official Beam documentation.