Datasets

Datasets

Every run appends what it collected to a dataset. Datasets are the reason to run something on a schedule: the value is the accumulated history, not any single execution.

Where they are

Outputs lists datasets across the install. A workflow's own detail page shows just its data.

What you can do

• Browse — a grid with sorting, filtering, column sizing and full-screen.
• Search across everything — one query over all history, not per-run. Often the fastest way to answer a question is to search what you already collected rather than run anything.
• Export — pull the records out.
• Delete — individual records or in bulk.

Through the API and MCP

Surface	Tool / endpoint
MCP	writ_workflow_data, writ_search_data, writ_export_data
REST	the dataset endpoints, with a key holding datasets:read

See MCP and REST API.

Retention

Run records are purged after RUN_RETENTION_DAYS (default 90). Set it to suit how far back you need to look — a long window on a busy install is the main thing that grows the database.

Files

Screenshots and downloaded files are stored on the data volume when WRIT_FILES_DIR is set — which the shipped compose does. Leave it set. Without it, binary objects are base64-encoded into the SQLite database and it grows quickly.