Scheduled agents

Use an agent for work you would otherwise ask for in the chat again and again: checking your sources for changes every Monday, refreshing a course when a policy moves, tidying module scripts each night. An agent is a name, a set of instructions and a schedule. Each time it runs, Vidext's agents follow those instructions on their own, the way they would in the chat, and nothing they change reaches your content until a person accepts it.

Agents live on the Agents page, under Notifications in the sidebar. Needs a look comes first, with an open card for each agent that is running, has changes waiting for review, or failed its last run: what it is doing or what went wrong, a strip of its latest runs, how its last two runs ended, and what you can do next. On schedule shows every other agent as a card drawn like the templates: its name, the start of its instructions, how often it runs and when it runs next. Point at a card for Run now and Pause; a paused agent offers Resume, and says why when it paused after failed runs. Add another offers the templates below them. Admins and members can create, run, edit and review them; learners never see them.

Create an agentCopy Anchor Link

On the Agents page, choose New agent. A new conversation opens on Home with a request for an agent, and Vidext's agents ask you what it should do, which courses and sources it covers, and when it should run. Then they propose it on a card, as in From the chat: nothing is created until you choose Schedule there. If you had already started writing a message on Home, it stays as it is and nothing is sent for you.

To start from a ready-made agent, such as Policy change watch or Translation follow-up, pick it under Add another. Every template is listed, each with what it does and when it would run. Picking one opens the same kind of conversation with the template's instructions and schedule already in the request, so the agents can propose it straight away, or first ask what they still need. A template that runs only when you run it leaves the schedule to you: the agents ask when it should run, and the card's Edit offers Manual only. A page with no agents yet shows a line on what agents do and the templates, under Start from a template.

Change an agentCopy Anchor Link

Edit, on an agent's card or in its panel, opens a window over the page with the agent's name, instructions and schedule.

• Name and instructions. Write what it should do each time, as you would ask in the chat. It can read everything your organization has: its courses, modules and sources.
• When it runs. Choose Manual only, Hourly, Daily, Weekdays, Weekly or Custom, then the time, the day and the Time zone the schedule follows. Custom takes a five-field cron expression, such as 0 9 * * 1-5 for nine on weekdays. Under the schedule, a line says when the agent runs next and in which time zone, that it runs only when someone presses Run now, or that the schedule never runs.

Choose Save, or Cancel to leave it as it was. Saving makes you the person the agent runs as; see Who it runs as.

An agent runs at most once an hour: neither Edit nor the chat accepts a schedule that would run more often.

From the chatCopy Anchor Link

You can also ask for an agent in the chat: say what should happen and how often, for example "Every Monday at nine, check our safety sources for changes and update the modules that teach them." Vidext's agents propose one, and a card takes the place of the message field with its name, its schedule in words, its time zone, when it would run next, and its instructions, written so a run can follow them alone: what to check and against what, what justifies a change, what stays as it is, and what a run reports when nothing is needed. Read them before you schedule: they decide how often a run asks you to review something (Show less folds long ones).

• Schedule creates the agent as you, in your time zone, switched on. The conversation then says Scheduled with the agent's name and schedule, and Open takes you to it on the Agents page.
• Edit opens the same fields as an agent's Edit, so you can change the name, the instructions, the schedule or the time zone first. If the suggested schedule can't be used, for example because it would run more than once an hour, the card opens on these fields and says so.
• Not now leaves it unmade. You can also type an answer instead, such as another day; the agents reply, and can propose it again with your change.

Nothing is created until you choose Schedule. From then on the agent works like any other: it is among your agents on the Agents page, it runs as you, and every change it makes waits for review, with its later scheduled times skipped until you review it. It counts toward your organization's five active agents: when all five are in use, Schedule creates nothing and says so.

Not available from the chat: one-time reminders, or a run on a single date. An agent repeats on its schedule.

Who it runs asCopy Anchor Link

A scheduled run works as the person the agent runs as, with what that person can do in your organization. That is the person who created the agent, or whoever edited it last: saving an agent's changes makes you the person it runs as. Resuming an agent that paused because its person lost access also makes you that person. Run now works as whoever presses it. The agent's panel shows who its schedule runs as under Runs as.

Each run works in a new conversation of the person it runs as. That person can open it from the run with Open conversation to see what the agents did and why; nobody else can.

The run's conversation is read-only, from the moment the run starts: you can follow it and read it, but not write in it, so the changes you review are only ever the run's own. To keep working on what it made, choose Continue in a new conversation there: the new conversation refers to the run's.

What a run doesCopy Anchor Link

A run reads your organization's content and makes its changes the way the chat would, with these differences:

• Every change waits for review. The run's changes are held together as one set. Nothing reaches your courses or modules until someone accepts it.
• Nobody answers its questions. Where the agents would ask you, they take the option they recommend and say so in the run's summary.
• Some steps are left to you. A run doesn't build modules, change a course's style, draw a course's cover, or make or remove presenters. Do those in the chat once its changes are in.
• It uses your organization's tokens, up to 2,500 per run. If your organization is out of tokens, or its plan payment is overdue, the run is skipped. A run that has used 2,500 tokens stops there, however many your organization has left. Calls already under way when it reached the limit still count, so a run can end a little over it.
• It has two hours and a quarter. A run still working after that stops.
• It reads only sources already connected. A run can search and read the sources that the person it runs as has connected, but it can't connect a new one: connecting needs someone there to answer.

When the run ends, its panel shows a short summary of what it did. A run that stops at either limit fails: nothing it changed is offered for review, though its work so far stays in its conversation. Like any failed run, it counts toward the three failures in a row that pause a scheduled agent.

Review its changesCopy Anchor Link

An agent with changes waiting moves to Needs a look, and its card says how many changes wait. Review opens them inside the card, one row for each run that left changes: when it ran, how many files it changes in how many courses, and the start of what the run says it did, or Proposed changes when it said nothing about them.

• Accept applies that run's changes to your content.
• Decline discards them. Declining asks you to confirm.
• See changes opens the full review: each file the run added, changed or removed, with its lines before and after, and Accept and Decline there too.
• Later closes the review and leaves the changes waiting.

The review closes by itself once nothing is left to decide. A run's changes also open from the agent's panel: under Changes, choose a set marked To review. A long run summary in the panel opens folded: Show all reads the rest.

Changes you accepted that couldn't be applied stay in the card, with Decline and See changes: the full review says why and offers Try again when trying again can help. In the panel they show Couldn't apply.

Changes from an archived agent are reviewed under Archived, on that agent's card. While any wait, a line on the Agents page points you there.

Accepted changes appear as Applying for a moment, then Accepted. Only the changes you reviewed are applied, and all of them or none:

• If the run's conversation changed after its changes were set aside for review, the review shows what it holds now and asks you to review it again.
• If someone changed some of the same files since the run made them, or some of its files can't be saved to your content, nothing is applied and the changes show Couldn't apply. Decline them and run the agent again.
• If the changes were taken back in the run's conversation since, there's nothing left to apply: decline them.
• If they couldn't be applied for another reason, choose Try again, or decline them.

If the changes were updated while you had the review open, Accept applies nothing and asks you to review them again.

Accepting changes a module's script, not what its learners play. When the changes touched a script, the confirmation says so and Open in Home takes you to that module in the conversation its course is written in, or a new one, to build it again. Learners keep the version they have until you build the module and publish its course.

Run, pause and archiveCopy Anchor Link

• Run now starts a run at once, from the agent's card or its menu, or from its panel. An agent runs one run at a time: while one is on its way, you can't start another with Run now, and a scheduled time that comes up meanwhile is skipped.
• One set of changes at a time. While a run's changes wait for review, or wait for you to try again or decline them after they couldn't be applied, the agent's next scheduled times are skipped, so its changes don't pile up on the same files; Run now still starts a run. The schedule picks up again once those changes are accepted or declined.
• Five active agents per organization. An agent that is switched on and has a schedule takes one of them; a paused agent keeps its slot, and a Manual-only or archived one takes none. At five, creating another, switching one on, restoring one or giving one a schedule is refused, and the organization keeps the agents it has until one of them gives its slot back.
• Three runs at once per organization. It counts every run queued, starting or running across all your agents. At three, Run now is refused, and a scheduled time that comes up waits: it starts as soon as one of the runs ends, and its card says Waiting to start meanwhile. So agents you schedule together, say every night at two, all run, one after another. If the agent's next time comes while it waits, the run that starts is for that later time, since missed times aren't made up. A time that still can't start after six hours is skipped and recorded.
• Pause, on a scheduled agent's card or in its menu, stops its schedule, and Resume starts it again. Run now works either way.
• Paused. An agent pauses itself when the person it runs as can no longer build in your organization, or when a scheduled run fails and it is at least the third failed run in a row. Runs from Run now count toward the three, but only a scheduled run's failure pauses the agent; a run that succeeds starts the count again. Its panel says why. Resume switches it on again; if it paused because its person lost access, it then runs as you.
• Archive, from the agent's menu, stops it running. Its runs and changes stay. Find it under Archived, a link at the foot of the page, and choose Restore to bring it back to its schedule. Changes it left waiting are reviewed there, from its card.
• Missed times are not caught up. A run skipped or missed while an agent was off is not made up later; the agent waits for its next time.

Hear about changes, failed runs and pausesCopy Anchor Link

When a run leaves changes waiting, the person it ran as finds them under Needs you, at the top of Notifications: what the run changed, the agent's name and how many files, with Review, which opens the agent's review. They stay there until someone accepts or declines them. When an agent pauses, its owner is told; if it paused because the person it runs as no longer has authoring access, the organization's admins are told instead, since the owner can no longer receive it and they can resume it. When a scheduled run fails, or is skipped because your organization is out of tokens or its plan payment is overdue, the person it runs as is told: Agent name didn't complete its scheduled run, with why. Running out of tokens and an overdue payment are each told once, not again for each run until tokens are added or the plan is paid. A run that pauses its agent is told as the pause instead, and a Run now that fails isn't told, since you were there to see it. All of them open the agent.

They are Agents notifications, listed with the rest, and Agents in Notification preferences turns them all off. Their email is off until you switch on Email for Agents in the same preferences; from then on each of them also arrives by email. See Notifications.

Run historyCopy Anchor Link

The agent's panel lists its latest runs under Runs: whether each one ran on schedule or from Run now, when it started and ended, its summary, and its state: Queued, Starting, Running, Succeeded, Failed or Skipped. A run that failed or was skipped says why, from the limits and rules above.

Not available: accepting only some of a run's files, editing a run's changes before accepting them, and publishing. Publish a course yourself once its changes are in; see Publish a course.