...Agents can review literature, propose hypotheses, modify code, launch experiments, inspect results, and generate research artifacts. Experiments are tracked through a Git-native tree that preserves lineage and immutable snapshots. Logs, diffs, files, results, and other evidence remain attached to the work that produced them. Compute can run locally, over SSH, or on platforms such as Slurm, Kubernetes, Ray, Modal, and Hugging Face Jobs. Projects and experiment data remain local by default through a local dashboard and SQLite store.