Advanced Features
Fuzzball does more with a workflow than start it, stop it, and show you its logs. This section covers reaching into a running job, sharing workflows with your group, and running long-lived services – exposing them as network endpoints, scaling them with load, and serving AI models that clients outside the cluster can call.
These pages do not build on one another. Read the one that matches what you need:
- Getting an Interactive Shell and Executing Commands – open a shell or run a command inside a container while its job is still running.
- Interacting with Other Users’ Workflows – check on workflows started by others in your group, and add owners so they can act on yours.
- Job Hostnames – how jobs and services address each other by name inside a workflow, and why those names have to be unique.
- Workflow Endpoints – list the live endpoints a workflow service exposes, inspect one, and mint an access token for an endpoint.
- Connecting to Workflow Services – reach a running service with
fuzzball workflow connector the Connect button in the web UI. - Service Autoscaling – run a service as a pool of replicas that grows and shrinks with load. Early access.
- In-Workflow API Access – call the Fuzzball API from inside a job using the token injected into every container.
- Serving an AI Model – start a large language model on a cluster GPU and get a base URL that speaks the OpenAI API.
- Using OpenCode with a Fuzzball-Hosted Model – point the OpenCode coding agent at a model you are serving on Fuzzball.