Data

Add data when you first need it so the thread records when and why it entered the research.

In the composer, open Add files, code, models to:

• upload files from your computer
• upload from a connected cluster
• browse existing resources or files by location
• add a Hugging Face dataset

In the message, identify the dataset's role and any relevant revision, split, or path.

Private or gated Hugging Face resources require a connected token. Files on linked compute require access to their configured location. See Start a thread for all supported context sources.

Data on compute

On-demand jobs start with an isolated filesystem and receive the inputs declared for that job. Linked-cluster jobs use the data mounts configured for that cluster. See How files move through compute before a workflow depends on files produced by an earlier job.

Stored experiment files

Personal settings → Storage groups experiment files by organization, folder, thread, and experiment. Storage cleanup is permanent, and deleting files can affect later report inspection or follow-up work.