Skip to content
Open Aghata

Work

Tasks and the Execution Cockpit

Every piece of work is an inspectable execution session — here's how to read one.

Everything Aghata does runs as a task backed by an execution session, and every session can be opened in the Execution Cockpit. Chat requests, emailed goals, schedules, routines — different doorways, same session, same controls.

  • Origin — where the task came from: a chat thread, an inbox email, a schedule, a routine, a table record. Sessions link back to their origin so you always have the context.
  • The plan — what Aghata intends, visible before and during the run.
  • Steps — each tool call, browser action, or code run as it executes, with its result or error inline.
  • Pauses — approval, input, and authentication cards, each stating exactly what’s needed to continue.
  • Outputs — deliverables produced so far, linked into Library.
  • Usage — what the run consumed, so costs are never a surprise.

You don’t have to wait for a pause to act:

  • Guide — reply in the session mid-run; guidance lands between steps.
  • Take over — in Browser mode, interact with the live view directly.
  • Stop — halt a session; partial outputs stay in Library, marked as partial.
State Meaning Your move
Running Steps are executing Watch, or walk away
Needs approval A risky action is queued Approve or reject on the card
Needs input A question only you can answer Answer in the session
Needs authentication A site wants a human login Log in via live view
Failed Couldn’t proceed, told you why Read the last step, fix, retry
Done Deliverable in Library Review the output and its sources

Paused tasks wait — hours or days. Waking them costs nothing; they resume from where they stopped, not from the beginning.

When a session ends, anything worth keeping is proposed as memory with the session attached as evidence. Approve it and future tasks start from it — provenance included.