Jobs & apps

A run executes one command on a fresh GPU container, streams its logs, and records the exit code and cost. Runs are grouped into apps: by default the script's name, or whatever you pass with --app.

Run a script

ts run train.py --gpu h100 --image pytorch --app llama-ft --secret huggingface --volume checkpoints
ts run pipeline.sh --gpu l4 --cmd "bash pipeline.sh --steps 10"   # custom entrypoint
ts run train.py --gpu h100 --detach                                  # return immediately

The script (up to 64 KB) is written to /workspace before the entrypoint runs. Ship larger inputs through a volume.

Apps

The Apps page lists every app in the current environment with its state, sandboxes, runs, cost and last activity. Open one to see its runs, sandboxes and the combined logs of the latest runs. Deleting an app removes its records and stops anything still running.

Timeouts and cancellation

Runs default to a one hour timeout (--timeout, max 24). Cancel from the console or with the API. A run that exits non-zero is marked failed and the exit code is shown.