Skip to main content

Tasks

A task is work an agent does on its own — on a schedule, after a delay, or on demand. Tasks are how OtoDock goes from "answer my question" to "keep this running for me": a daily report, an hourly check, a weekly cleanup, a one-off job kicked off in the background.

A task run is just a chat

A task is the same kind of session as a chat — you can open a running task and watch it live (streaming text, reasoning, tools, file edits) exactly as you would a conversation. The task view and the chat view are essentially the same screen.

Creating a task​

Creating a task is as simple as asking. Tell an agent in chat — "summarize my open issues every weekday at 9am" or "remind me in 30 minutes" — and it sets the task up for you. Every task then appears in the agent's settings, where you can manage it.

Tasks come in three shapes:

  • Recurring — runs on a repeating schedule. Use a familiar cron schedule ("every weekday at 9am") or a plain interval ("every 6 hours"). Recurring tasks remember your timezone, so they keep firing at the right local time.
  • One-time — runs once, at a specific date and time or after a delay.
  • Trigger-fired — no schedule at all; it runs when an external event arrives through a trigger wired to it.

When several tasks share a fire time (a fleet of 6:30am briefings, say), OtoDock spaces their session starts a few seconds apart so they don't all wake at once. The fire times themselves stay exact.

A task can run on its own model​

Every run uses the agent's default model unless the task says otherwise. An agent can pin a single task to a specific model and engine, so an agent kept on a cheap model can still run its one demanding daily briefing on a stronger one (or the reverse) without changing anything else it does. Ask the agent to pin, retune, or clear it; both Scheduled Tasks pages show what each task will actually run on, with pinned tasks clearly marked. Agents are told to ask before pinning — which model a schedule burns is your spend.

Personal or shared​

Like agents themselves, tasks can be private or shared:

  • Personal task — only you see it, its results go to you, and its files stay in your private space.
  • Shared task — everyone with access to the agent sees it, and its results land in the agent's shared workspace. Managers and editors create these for the team. Contributors can't, because a shared task runs as the agent and a contributor only adds files. Viewers and contributors can watch a shared task's runs, but they can't type into them or stop them.

Runs and history​

Every time a task runs, it records a run: status (completed, failed, cancelled, or stopped at a usage limit), start and end time, duration, cost, and the output. Browse an agent's run history to see what it did, review results, or track down a problem. Runs can be renamed and, with the right role, deleted from the history. A run that dies on a real engine error (an expired login, a provider rejection) is reported as failed with the provider's message, never as a quietly empty success.

Checks on a task​

A task can name checks that review every run it starts. The agent gets as many rounds as the check allows to fix what it finds, and if the check still fails after that, the run is reported as failed, together with the findings.

Pick up where a task left off​

A finished task is also a conversation. Reopen any run and send a follow-up — the agent resumes with full context from that run, files and all, so you can refine the result or dig deeper without starting over. Once a run is over you can even continue it on a different model or engine from the chat's model picker, like any other finished chat.

Delegation: agents helping agents​

A task can hand part of its work to another agent that's better suited to it, and fold the result back in. You control which agents are allowed to delegate to which others — see Delegation.

Notifications when a task finishes​

Each task decides how it tells you it's done:

  • Automatic — you get a "Task complete" (or "Task failed") notification with a link straight to the result.
  • Custom — the agent writes its own notification with a message and severity it chooses.
  • Silent — it just runs in the background with no alert.

See Notifications for how alerts reach you.

Managing tasks​

  • Run, pause, resume, or delete any task from the agent's settings.
  • Edit a task's prompt, schedule, or notification mode by asking the agent — no need to recreate it.

How many tasks can be active​

So that no single person or agent floods the schedule, OtoDock limits how many scheduled and one-time tasks can be active at once:

  • 25 personal tasks per person, across all of their agents. Admins have no limit.
  • 100 tasks per agent, counting everyone's tasks together.

Paused tasks, trigger-fired tasks and tasks installed by an agent template don't count. When you reach a limit, creating or resuming another task is refused with a message that names the limit, so pause or delete a task first. An admin can raise either number in .env (TASK_MAX_ACTIVE_PER_USER, TASK_MAX_ACTIVE_PER_AGENT), or set it to 0 to remove the limit.

Next steps​