See scheduled tasks
Find the scheduled and recurring AI tasks people in your organization run, how often they run, and what each run costs.
Scheduled tasks are AI tasks that run on their own, such as a daily inbox summary in Claude Cowork or a Claude Code /loop. The Scheduled tab on the Sessions page lists them, with who owns each one, how often it runs, and what a run costs.
Who can see this: Admins. See Roles and permissions.
Open the list
Go to Sessions and click the Scheduled tab.
The first time anyone in your organization opens the tab, Tokenize searches your captured sessions for scheduled tasks. You see "Finding scheduled tasks" while it runs. This can take a few minutes.
Tasks come from these platforms: CoWork, Codex, Claude Code, Claude Code cloud, ChatGPT, and Cursor. Claude Code includes cron tasks and /loop runs on people's own computers, and tasks scheduled in the Claude Code desktop app.
Filter the list
Once the list has loaded, use the menus in the top right:
- All platforms, or one platform.
- All tasks, Recurring, One-time, or Short check-ins. A short check-in is a Codex task that checks on something often for a short while, such as watching a build or a pull request.
- Last 12 months or Last 30 days. This sets which tasks show and which runs the run counts and costs cover.
Refresh rebuilds the list from data Tokenize has already received. It doesn't ask devices or providers for anything new.
Read a row
Each row reads like a sentence: the owner, then "runs", "ran", or "will run", then the task name. Badges follow:
- Recurring, One-time, or Short check-in. Hover over it to see the schedule.
- A status, when the task isn't simply enabled.
- Last run: Failed, or another status, when the newest run didn't finish as intended.
On the right are the run count and the average cost per run, such as "12 Runs | $0.84/run". "(partial)" means some runs couldn't be priced. "(est.)" after a run count means some runs of a /loop were worked out from the loop's own timing. The Tokenize app reports them directly after its next update.
Tasks that people saved but haven't run yet also show. Click the owner's name to open their profile.
Task details
Click a task name to open its details:
- Owner, Platform, and Schedule.
- Status and Next run, when known.
- Last seen, the last time Tokenize saw the task.
- Settings checked, when Tokenize last read the task's saved settings, and when they last changed.
- Runs, with a count by outcome, such as "7 no update, 1 completed". "Hasn't run yet" means the task has never run. "Still collecting older runs, so this may go up." means Tokenize is still reading the task's history. "Some runs from the last year couldn't be collected, so this is low." means part of the history couldn't be read.
- Latest run or Runs in, a link to the session, or to the ChatGPT chat the task runs in.
- Cost per run, the average, median, and last run.
- Latest runs, each with its outcome, when it ran, and how long it took. ChatGPT runs link to their Chat. Codex runs link to their Session.
- Prompt, the instructions the task runs with. "No prompt captured." means Tokenize didn't see one.
With Last 30 days selected, Runs and Cost per run cover only that window.
Cursor Grok Bots
Grok Bots don't appear in the task list. Cursor bills the bot, not each routine, so they have their own Bots tab next to Scheduled. See See Cursor Grok Bots.
Troubleshooting
"No scheduled tasks found" with "No scheduled tasks showed up in captured sessions." Tokenize didn't find any scheduled tasks. Check that the desktop app is running on the computers where people schedule tasks. See Troubleshoot the desktop app.
"No matching tasks". Your filters hide every task. Try a different filter.
"Couldn't find scheduled tasks" or "Couldn't start the search". Click Try again.
"Last refresh failed. Showing earlier results." The list is from an earlier search. Click Refresh to try again.
"N task lists couldn't be read. Some scheduled tasks may be missing." Some tasks may not show. The rest of the list is accurate.
"Run history unavailable" on a ChatGPT task. ChatGPT didn't let Tokenize read that task's runs, so zero runs doesn't mean it never ran.
"Not priced. ChatGPT doesn't report token counts." ChatGPT runs are counted but have no cost.
"Not priced. Costs for local Codex runs aren't shown yet." Runs of Codex tasks on people's own computers are counted but have no cost. A Codex run's outcome is Completed when its turn ended, or Interrupted when it was stopped. A run that hasn't ended yet shows as Started. One whose end was never recorded shows as No end recorded.
Related
- Cards about costly or stuck scheduled tasks appear on the Opportunities board.
- For other sessions, see Browse sessions and workflows.