Skip to main content
Version: 0.8.4

Create and run a Work

Work is available to administrators and Advanced Users. The User role has no Work configuration or execution entry; use Chat for an interactive conversation.

Create something you can verify​

  1. Enter the target Workspace and create a Work, or ask Uno to help.
  2. Choose the responsible Agent, a recognizable name and a clear objective.
  3. Specify input files, processing requirements, output paths, completion criteria and when human confirmation is required.
  4. For a manual-only trial, choose a one-time Work and leave its scheduled time empty. Save and try a small, non-sensitive input.
  5. Check the final reply, output files and execution steps before enabling automatic triggers.

“Read this week's sales sheet, save a report under reports/, and ask about missing figures instead of guessing” is easier to verify than “analyze the business weekly.”

Choose a trigger​

  • Manual: a person decides when to start; suitable for trials or irregular work.
  • Scheduled: a one-time Work may have one planned start; recurring work uses its configured schedule. Check the working timezone and next planned run, rather than assuming your browser's timezone.
  • File-triggered: runs when configured file rules match. Check the watched location, matching scope and when input writing finishes; verify with a test file first.

Automatic work needs a running Engine and Scheduler. Quitting a Desktop host app, sleeping or shutting down its computer can prevent scheduled execution. Consider Linux for continuous unattended operation. Closing a browser tab does not stop a remote task.

Pausing a Work pauses automatic scheduling; it does not mean a running task has stopped, and manual execution remains available. Use the individual Run's Stop action when needed. A Work's finished schedule does not prove its last Run succeeded. The trigger type is read-only when editing an existing Work; create another Work if you need a different trigger type.

Inspect individual runs​

A Work can produce many Runs, each with independent history. Open the relevant run to inspect state, Agent steps, final reply and recorded usage. Attention may indicate a pending human question. On failure, read the failing step and error before changing configuration or rerunning.

Stopping or rerunning does not undo files, sent messages or external actions. Plan how repeated execution will avoid duplicate submissions before automating changes to another system.

Automatic runs check Workspace access using the task creator's identity; manual runs use the person starting them. Revoked access, an inactive Agent or unavailable models/tools can block later runs. Delegating to an Agent is not a way around user access rules.

Related guides: human decisions, roles and access, file workbench.