Managing it

Everyday commands

docker logs -f neurogrid-agent      # follow the logs
docker stop neurogrid-agent         # stop serving, stay registered
docker start neurogrid-agent        # start again
docker restart neurogrid-agent      # stop and start
docker ps --filter name=neurogrid   # what is running right now

Stopping is not unregistering. Your GPU shows as offline in the console and comes back when you start it again.

Is it healthy

docker logs --tail 20 neurogrid-agent

A healthy agent prints Heartbeat sent successfully every 30 seconds. If you see gRPC error sending heartbeat repeating instead, go to Troubleshooting.

Am I running a job

docker ps --filter "name=neurogrid-job" --format '{{.Names}}\t{{.Status}}'

Job containers are named neurogrid-job-<deployment id>. Empty output means your card is idle.

Taking a break

Pause the listing in the console rather than stopping the container. The backend then stops offering you work, and any job already running finishes normally instead of failing.

Stopping the container mid-job fails that job, which counts against you.

Removing it

docker stop neurogrid-agent
docker rm neurogrid-agent

This keeps the neurogrid-data volume, so the machine stays registered and reinstalling picks up where it left off.

To remove it completely, including the registration:

docker volume rm neurogrid-data

After that the machine needs a fresh token from the console to come back. Also cancel the listing there, or it will sit as an offline GPU indefinitely.

Disk the agent uses

What	Roughly	Safe to delete
Agent image	250 MB	No
vLLM engine image	20 GB	Yes, but the next job re-downloads it
Model cache	grows with what you serve	Yes, same trade
neurogrid-data volume	~1 MB	No — this is your registration