Actions / Documentation
Run your first job, then look things up.
Two paths: the guided setup if you have an approved repository and a device to run jobs on, and the reference if a job behaves in a way you did not expect. Everything here describes the beta as it ships, including what it deliberately does not do.
- Start here
Install and enroll a device
The guided setup: prerequisites, connecting an approved repository, running the installer, and starting a bounded session that takes jobs.
- Prerequisites
- Connect a repository
- Run guided setup
- Start a session
- Concepts
How Tbench Actions works
The model behind the one-line change: requests, the queue, device leases, the lifecycle of a job, and the boundary between GitHub and Tbench.
- The model
- Queue and leases
- A job’s life
- Credential flow
- Isolation
- Reference
Labels, capacity and commands
Look-up material: the runner labels and their CPU and memory, device capacity flags, environment variables, egress profiles, worker commands and local files.
- Runner labels
- Device capacity
- Environment variables
- Egress profiles
- Worker commands
- Troubleshooting
When a device or job stops
What each failure means and what to do next: Docker stopped or too old, a job that never runs, a job sent to attention, blocked cleanup, and enrollment problems.
- Run the diagnostics
- Device won’t start
- Job isn’t running
- Cleanup and recovery
- Operators
Approve scope and issue access
The operator side: approving repository scope, issuing a single-use enrollment, setting workflow and branch policy, and reading the pilot console.
- Approve scope
- Issue an enrollment
- Set policy
- Read the console
Not finding it? The reference covers every environment variable and command the device honors, and the troubleshooting page names the exact error strings the worker prints. For anything else, write to the team.
Start with your actual workload.
Explore the model, then apply for a scoped pilot. No card or compute commitment.