A task’s record and cost
Every task keeps a full record: what it was asked, what it delivered, each step it took, what waited for you and what every call cost. Open it to understand what an employee did, and why it cost what it did.
On this page
Opening a task#
Select a task on Work, in an employee’s Work tab, or from Open the task in the Inbox. A short ID such as KS-142 in Chat links there too. The header shows the ID, status, what kind of task it is (Team task, “Part T2 of …”, Goal plan, Routine), the employee, when it started and how long it ran, and its cost. Copy link copies a link to it, and Pin to the rail keeps it one click away.
What the record holds#
| Section | What it tells you |
|---|---|
| Waiting for you | Only while something is pending: the question, approval, guidance or limit raise, with the controls to answer it. Nothing runs or is charged while it waits. |
| Summary | What the run did, in a few lines. |
| Deliverable | The document it produced (“Markdown · 640 words”) with Open in Artifacts, any Files it kept, and the Sources it cites, each opening in Knowledge. |
| Team plan and Done when | For team tasks and tasks held to a skill’s checks: each part or check and its verdict. |
| Assignment | Why this employee: you chose them, a goal’s plan gave it, or “Routing picked {name}: {reason}”. |
| Notes and direction, Revisions | What you told it along the way, and each revision you asked for. |
| Timeline | Every step, oldest first, across every run. |
| Details | Employee, model and reasoning, setup version, times, cost and attempts. |
Reading the timeline#
The timeline is the task’s history from Royak’s records, not the employee’s own account. A reload shows exactly the same entries. Hide steps leaves out the work’s own steps so you see only the milestones; Show steps brings them back.
| Entry | Means |
|---|---|
| Assigned | To whom, and whether by you, routing, a goal plan or a team plan. |
| Work started / Revision N started | A run began, with the employee, model and setup version it used. |
| Knowledge search | What it searched your sources for, and the passages it found. |
| Connection read / Web page | What it read from an app or the web, with the page’s title, address and a screenshot. |
| File, Kept file | A file it looked at, or kept from a result, shown in place. |
| Action | Something it wanted to do in an app: waiting for approval, or its outcome. |
| Question from {name} / Your answer | A question it asked you, and what you said. |
| Not making progress | Why it stopped, what it had spent, and Your guidance. |
| Approval requested / decided | An action you were asked to approve, and your decision. Limit raise requested and decided do the same for spending limits. |
| Outside content | It read something you didn’t write, marked Not written by you. |
| Finished | Done, Failed (with the reason) or Canceled, and what that run cost. |
A repeated action is never sent twice: the step reads “Already done earlier in this task; not sent again.” A repeated search is answered from the first one; nothing is searched again. A task with more than 50 runs shows its latest 50.
What it cost, and why#
A task’s Cost is every AI call of every run it made: its first run, each revision, and for a team the planning, review and combining calls. Each run’s share is on its Finished entry. Amounts read in credits while your plan’s credits pay, else in money.
| What drives cost | Why |
|---|---|
| The model | Stronger models cost more per call. See the employee’s Model in Overview. |
| The number of steps | Each step is one call. A run takes at most 8 model steps and 4 knowledge searches. |
| How much it reads | Long briefs, instructions and passages make every call larger. |
| Revisions and returns | Each revision, and each time a review or check returns the work, is a new run. |
| Reviews and judged checks | One call each, on the reviewer’s model. |
Details lists each attempt: valid reply, unusable reply, asked a question, outcome unknown, and so on. Each step may try twice with the same request when the first reply isn’t usable.
Steering a task while it runs#
- Note for {name} and Leave note: read at the next step, never by one already under way. A note can’t grant tools or skip approvals; Withdraw takes it back before it is read.
- Change direction: replace the brief. The current work stops at its next step and continues with the new brief; approvals it asked for and that weren’t sent are withdrawn.
- Reassign to: before work starts, move the task to another active employee.
- Pause and Resume, or Cancel task… to stop it for good.
- Request revision: once it has a deliverable; see Teams and handoffs.
Questions#
Why did a small task cost more than I expected?
Check the timeline: several steps, a large document read in full, a revision or a returned review each add calls. Details names the model.
Is the summary the employee’s own claim?
The summary is the run’s; the timeline, cost, checks and verdicts come from Royak’s records. A task completes on its deliverable and checks, never on an employee saying it is done.
Can I see what a page looked like when it was read?
Yes. A Web page entry keeps its screenshot; select it to open it full size.
What happens to work when I cancel?
A call under way finishes and its cost stays recorded. The task and any tasks waiting on it won’t continue.