How jobs run

Jobs do real work behind a page: a PDF of an invoice, a page for every row of a spreadsheet. You start them; visitors never can.

Ready-made jobs

A PDF export prints exactly what a browser shows, in a locked-down container: no network except the font and library servers a page needs, nothing it can keep, and a time and CPU limit. "From spreadsheet" makes one page per row of a CSV, from a template. No code you or an AI wrote runs on our servers.

Files

Files a job makes are kept for an hour, then deleted. Download them from the Jobs page, or through the API (GET /api/jobs/:id/files/:fid).

Limits

Each workspace has a daily budget of compute time and a short queue. A job that runs too long is stopped and its time counted.

Jobs on your own computer

Under Jobs → On your computer you can write your own commands. They run only in the desktop app on your computer, with the inputs you chose, after you approve each command there once. Nothing listens on your computer: the app asks for work.

From the API and AI tools

Start and check jobs with POST /api/jobs and GET /api/jobs/:id, or the run_job and job_status MCP tools. See the API and MCP reference.