Hugging Face Jobs runs your Python script or Docker command on CPU/GPU hardware you choose. You supply the environment and code; HF handles provisioning. CLI, Python and HTTP API let your own tools or agents submit work.

Hugging Face Jobs is a managed execution service: submit a command, choose an image and hardware, and follow the run's status and logs.

Python scripts can run through hf jobs uv run, which installs declared dependencies with uv. Docker Jobs use hf jobs run and can run other languages or custom environments. The same service is available through the huggingface_hub Python client and an HTTP API.

Typical workloads include training and fine-tuning, model evaluation, batch inference, and data processing. A custom agent runtime packaged in a suitable image is a possible workload too; that is an application of the execution model, not a claim that Jobs manages an agent's memory or conversation.

Example launch, shown for illustration and not executed:

hf jobs uv run --flavor t4-small --timeout 2h train.py

Source: Jobs overview.