Workrooms

A workroom holds the inputs, agents, plan, and output for one deliverable — not for an entire client engagement.

The 30-second version

A workroom is the unit of work in AtelyaOS. Each workroom has a goal, one or more deliverables, a task plan, system agents (Planner + Composer), optional custom agents, and a credit budget. Workrooms move through a fixed lifecycle and produce one or more exported deliverables.

Why this matters

Marketing-tool conventions push you toward "one project per client" and a flat chat. AtelyaOS pushes the other way: one workroom per deliverable. That keeps the Planner's context narrow, makes credit costs predictable, and means revisions only re-touch what you scoped, not three quarters of unrelated history.

A single client typically has many workrooms over time: a proposal, a weekly recap each Monday, a scope-change doc, a pitch outline. They share a workspace, an export profile, and any client-specific custom agent — but each is a clean piece of work.

How it works

Lifecycle

Every workroom moves through this state machine:

planning → executing → composing → completed
              ↓             ↓
            paused    awaiting_revision (if you request one)
              ↓             ↓
           cancelled    completed (after revision)

Terminal states are completed, failed, and cancelled. From a failed workroom, you can replan to retry.

What a workroom contains

• Goal — one paragraph describing the outcome.
• Deliverables — short labels (e.g. "Q4 proposal", "3 Instagram reel scripts").
• Constraints — optional deadline, max-credits cap, paste flags.
• Task plan — produced by the Planner, ordered, each task assigned to an agent or role.
• Agents — Planner and Composer (system) plus any custom agents you add.
• Style override — optional one-shot style instructions for this workroom only.
• History — every task event, message, and revision round, viewable in the activity tab.

[SCREENSHOT: workroom detail page showing task list and state badge]

Lifecycle in plain English

1. Planning. You start the workroom. The Planner LLM reads the goal, inputs, and available agents and produces a task plan.
2. Executing. Tasks run in dependency order. Some are agent work, some are tool calls; a few may pause for approval.
3. Composing. A final composition task assembles the deliverable.
4. Completed. You can review, export, or open a revision round.

The transition from planning to executing is automatic — there is no separate plan-approval gate today. If the plan is wrong, use Replan (which discards non-completed tasks and re-invokes the Planner) or Pause to stop the run.

Workroom vs. project

A workroom isn't a folder for ongoing work. It's closer to a "ticket" with a built-in plan, a credit meter, and a deliverable at the end. If you want one place to see everything for a client, group workrooms by client name in the title (see first workroom).

Common pitfalls

• Treating workrooms as long-running projects. A weekly report is one workroom per week, not one workroom for the year. The credit math and Planner quality both degrade with very long histories.
• Skipping inputs. A workroom with only a goal and no source material leans on LLM defaults. Paste the brief, the past sample, the meeting notes.
• Forgetting to export. The workroom is your staging area. Until you export and send the .docx / .pdf, nothing has shipped.
• Replanning in mid-execution. Replan stops in-flight tasks. Use Pause if you only want to think for a minute.

What's next

• Custom agents — beyond Planner and Composer.
• Briefs and plans — the inputs that drive a workroom.
• Approval and revisions — how the loop closes.