How can we help?

Articles in this section

See more

Orion Pulsars: agents that keep watch for you

Jeff Farris
  • Updated

An Orion Pulsar in CloudRadial AutomationAI is an agent that keeps working toward one goal for you. It checks in on its own, writes notes about what it finds, and proposes changes for you to approve before anything changes. This article is for workspace users of any role who create and look after Pulsars. Approving changes, and a few settings, need the Admin or Owner role.

What a Pulsar Is

A Pulsar pairs an agent with a one-sentence goal, such as "Keep every client's critical patches under 7 days old", and keeps that agent working on it in the background. Use a Pulsar when a job is ongoing rather than one-off: watching patch levels, alert queues, license counts, or ticket hygiene, and fixing what drifts.

Each check-in follows the same cycle:

  • It looks. The agent runs its read-only tools for real, so it sees the current state of your systems.
  • It proposes. Any tool call that would change something is not carried out. It becomes a proposal, with the exact tool, arguments, and the agent's reason.
  • You decide. The proposals from one check-in are collected into a batch for you to approve or deny.
  • It acts. Only the proposals that were approved are carried out, each exactly as approved.

The check-in itself never changes anything, whatever the Pulsar's settings. Along the way the Pulsar keeps notes, learns from your decisions and answers, and assesses whether it is reaching its goal.

Before You Start: A Runner and an Agent

Pulsars appear in the sidebar as Pulsars, in the Actions section above Playbooks, when the feature is available to your workspace. The entry shows a count of items that need attention.

A Pulsar needs two things:

If the runner is missing or no longer available:

  • With no runner installed, the create page shows "No runners are installed yet. Install one from Runners first — a Pulsar needs one to work on."
  • Saving without a runner is refused with "Choose the runner it works on."
  • If the runner is revoked or removed, the Pulsar halts at its next check-in. Nothing is sent to any runner, and the Halted banner reads, for example, "It stopped: its runner Main-Office is no longer available — choose another in Settings". Resume is refused until you choose another runner on the Settings tab.
  • If the runner is only offline, the check-in waits for it. Activity shows "Waiting for runner to pick it up. Check that runner is online."

Check-ins use the AI provider configured on the runner, which your own AI provider bills directly; they don't consume Usage Credits. See Choosing the runner AI provider.

Creating a Pulsar

Go to Pulsars and select New Pulsar. The page asks five questions. Above them, a summary sentence updates as you answer, so you can read back what the Pulsar will do.

  1. Who's doing the work? Choose the agent. Selecting one shows its description and which of its tools can change things, so you know what kinds of proposals it could make. Select New agent to build an agent in a new tab; it appears in the list when you come back. Then choose the Runner. The list shows each runner as online or offline; with only one runner, it is chosen for you. Optionally enter a Name; leave it blank to use the agent's name.
  2. What's the goal? Enter the Goal in one sentence. It is repeated to the agent at every check-in. Optionally fill in How will we know it's done?
  3. When should it check in? Choose On a schedule, When something arrives, or Keep going. See Choosing When It Checks In.
  4. How much can it do on its own? Choose one:
    • Watch and report: it looks and writes notes, but never proposes a change
    • Propose changes (the default): it asks before every change, and nothing changes until someone approves
    • Act on approved kinds: tools you choose run without asking, and it asks before anything else. Only Admins and Owners can choose this; see Letting It Act Without Asking.
  5. Where does it keep notes? Choose a new folder named after the Pulsar, or An existing folder. Notes are documents in Knowledge, and the Pulsar reads its recent notes at every check-in.

Expand Limits to see the Monthly usage limit (tokens) and Usage limit per check-in (tokens), prefilled with the workspace defaults, and two optional finish rules: Finish when it reports the goal reached at two check-ins in a row, and Finish on a date.

Before starting it, you can select Try a check-in now. This saves the draft and runs one check-in straight away without starting the schedule; the Pulsar stays a draft. When it finishes, you see its summary and counts, with links to Review proposals and See it in Activity. Anything it proposes waits for review like any other proposal.

Select Save draft to save. The Pulsar opens in Draft; select Start on its page to begin checking in.

Choosing When It Checks In

On a schedule checks in at set times. Choose Every hour on weekdays, Every weekday morning, Every day, Every week, or Custom, then set the time and, where it applies, the days. Times are in the workspace time zone, which is named on the page, and the next check-ins are previewed under Next check-ins.

When something arrives checks in each time another system sends a request to the Pulsar's own trigger address. See Starting Check-Ins From Another System.

Keep going starts the next check-in a set time after the last one finishes. Enter how many minutes to wait; the minimum is 15 minutes.

Whatever the trigger:

  • A check-in that comes due while the previous check-in, or approved changes, are still running is skipped. Activity records that it was skipped and why.
  • Proposals still waiting for review don't hold up the next check-in. The Pulsar is told what is still pending, so it doesn't propose the same thing again.
  • You can check in at any time with Check in now on the Pulsar's page.

Starting Check-Ins From Another System

With When something arrives, each POST to the Pulsar's trigger address starts a check-in. The request's JSON body is given to the Pulsar as context for that check-in (the first 4,000 characters), so an alert or ticket can tell it what to look at. A request that arrives during a check-in waits its turn.

The trigger address and its secret are created when you first save:

  • After Save draft, the Pulsar opens on its Settings tab with the Trigger address and Secret shown, each with a copy button.
  • The secret is shown only once. Copy it now and send it in the X-Crauto-Webhook-Secret header of every request.
  • Later, the page reads "A secret is set. It is only shown when generated." If you no longer have it, select Generate a new secret and update the system that calls the Pulsar.
  • The trigger address also appears on the Overview tab, under Next check-in.

To test the trigger before wiring up a system, expand Test it on the Settings tab. It has ready-to-copy requests With curl (macOS, Linux) and With PowerShell (Windows), filled in with the trigger address and, while it is on screen, the secret. They look like this:

curl -i -X POST "<trigger address>" -H "Content-Type: application/json" -H "X-Crauto-Webhook-Secret: <your secret>" -d '{"message":"Test from curl: check the latest alerts"}'
Invoke-RestMethod -Method Post -Uri "<trigger address>" -Headers @{ "X-Crauto-Webhook-Secret" = "<your secret>" } -ContentType "application/json" -Body '{"message":"Test from PowerShell: check the latest alerts"}'

The response tells you what happened:

Response Meaning
202 A check-in started or is queued
409 The Pulsar isn't running. A draft or paused Pulsar answers "This Pulsar isn't running. Start it, then send the request again."
404 The address or the secret is wrong

The Pulsars Page and a Pulsar's Workspace

The Pulsars page opens with Needs you: proposal batches to review, questions to answer, and halted Pulsars to resume, oldest first, each with one action (Review, an answer box, or Resume). Below it, All Pulsars lists every Pulsar with its agent and runner, status, goal, last and next check-in, pending proposals, and usage. Filter by status, or select Needs attention.

A Pulsar's status is one of Draft, Active, Checking in, Waiting for you, Paused, Halted, or Done.

Select a Pulsar to open its workspace. The header shows its agent, runner, and goal, and these actions:

  • Start, Pause, or Resume, depending on its status. Pausing stops new check-ins; proposals waiting for review stay reviewable, but approved changes aren't carried out until you resume.
  • Check in now. On a draft, this runs a trial check-in. A paused or halted Pulsar must be resumed first.
  • More: Edit (opens Settings), Duplicate, and Delete. A duplicate is named "Copy of …", gets its own notes folder, and never inherits permission to act without asking.

The workspace has five tabs:

  • Overview: the goal, the Latest assessment in the agent's words (Goal reached, Improving, or Not improving), the Last check-ins, what needs you, the Next check-in, and this month's Usage
  • Activity: every check-in, newest first
  • Proposals: every proposal batch, open ones first, with the pending count on the tab
  • Memory: notes, what it has learned, and its next briefing
  • Settings: the same five questions as creation, plus Who should hear from it?, Limits, and Stop or delete

Under Stop or delete, Stop for good ends the Pulsar: it won't check in again, and its activity and notes are kept. Delete Pulsar removes the Pulsar, its activity, its proposals, and what it has learned. Either way, the notes it wrote stay in Knowledge.

Following Check-Ins in Activity

Each entry on the Activity tab shows the check-in number, what started it (schedule, event, manual, trial, or keep going), its status, start time and duration, the agent's summary, and what it looked at, proposed, noted, and asked. Expand an entry to see:

  • What it was told: the exact briefing the agent received
  • Steps: each tool call, with changes shown as proposed, not done
  • Notes written, Proposals, and Approved actions carried out

Skipped check-ins appear as single lines with the reason, and actions such as starting, pausing, or resuming are listed with who took them.

Each check-in names its runner and when it was picked up. While one is running, Activity explains a delay in plain words:

  • A check-in no runner has picked up after two minutes reads "Waiting for runner to pick it up."
  • A check-in with no word for five minutes reads "No word from runner since …", with the time by which it will be stopped if the runner hasn't reported back.
  • A check-in that timed out explains the likely cause: usually a tool returned more data than the runner could hold, or the runner restarted. Try a narrower goal or filter, and check the runner's logs if it keeps happening.

Reviewing Proposals

When a check-in proposes changes, they are collected into one batch. You can open a batch from Needs you, the Proposals tab, a notification, or the Inbox. Each proposal card shows a plain title, the target (such as a company or device), the tool, a risk label, the agent's reason under Why, and the exact arguments under What it will send.

For each pending proposal:

  • Approve carries it out as shown
  • Deny rejects it, with an optional reason. The Pulsar remembers the reason and uses it at later check-ins.
  • Edit lets you change the tool's arguments in a form, shows the before and after, and then Approve with changes. Edits are checked against the tool's own rules.

To decide the rest of a batch at once, use Approve all remaining or Deny all remaining. High-risk proposals and tools that always ask are never approved in bulk; they are left for individual review. Keyboard shortcuts are available: J and K move between proposals, A approves, D denies, E edits, and ? lists them.

When every proposal in the batch is decided, the approved ones run on the Pulsar's runner. Select Run approved now to start them without waiting for the rest. Until they start, you can Undo a decision. Each proposal then shows Done with its result or Failed with its error; one failure doesn't stop the others. If the agent tries a change that wasn't approved while carrying them out, the change isn't made and becomes a new proposal.

A few more rules:

  • Proposals expire if no one decides them, after 72 hours by default. The batch shows Expires in, then Expires soon in its last 12 hours. An expired proposal can no longer be approved.
  • A check-in proposes at most 20 changes by default; the batch notes how many more were dropped.
  • If a check-in proposes a change that is already waiting for review, it isn't added twice. The card shows how many times it was proposed.

Pulsar batches also appear in the Inbox. A batch with a single proposal can be approved or denied right there; Review opens the full review. Questions from Pulsars appear there too, with an answer box. See Handling Action Required items in the Inbox.

Letting It Act Without Asking

With Act on approved kinds, an Admin or Owner chooses tools whose proposals are approved automatically:

  1. Under How much can it do on its own?, choose Act on approved kinds.
  2. Under Which changes can it make without asking?, select the tools. Tools marked always asks can't be selected; they always need a person to approve them.
  3. Select Allow these…, check the tools named in the Let it act on its own? dialog, and confirm.
  4. Save the Pulsar.

The check-in still only proposes. Proposals for allowed tools are approved automatically and carried out like any approved change; everything else still waits for review. Auto-approved proposals stay on record, labeled Auto-approved with the name of the person who allowed the tool.

Memory: Notes, Learnings, and Questions

Every check-in starts with a briefing: the goal, the latest check-in's summary, current notes, what it has learned, and relevant knowledge. The Memory tab shows all of it:

  • Notes: what the Pulsar wrote, each linked to the check-in that wrote it. Notes are documents in the Pulsar's notes folder; select the link to open the folder in Knowledge. When a newer note replaces an older one, or old notes are combined into a summary to keep the briefing short, the originals stay visible under Show superseded and Show compacted.
  • What it has learned: preferences from your decisions (every denial, and every decision with a reason) and answers to its questions. Any member can Edit or Delete an entry; a deleted entry is no longer used.
  • Next briefing: exactly what the agent will be told at its next check-in, with its length against the limit

A Pulsar can also ask questions instead of guessing. A question appears in Needs you and the Inbox; type your answer and select Answer. The answer is saved to what it has learned, and the Pulsar uses it from the next check-in. Unanswered questions are not asked twice. For more on folders and documents, see Managing your knowledge base.

Halts and Limits

A Pulsar stops itself, and shows Halted, when something needs a person. A banner on its page says why and offers the fix your role allows:

Why it halted What to do
It reached its monthly usage limit An Admin or Owner selects Raise limit; then Resume. Raising the limit alone doesn't restart it.
Two check-ins in a row went over the usage limit per check-in Raise the limit or narrow the goal, then Resume. The proposals from those check-ins are still collected for review.
Its last three check-ins failed Check Activity for what went wrong, then Resume
The agent it uses was removed Choose another agent on Settings
Its runner is no longer available Choose another runner on Settings, then Resume

Usage is counted in AI tokens the Pulsar's check-ins and approved changes use. The Usage meter on Overview reads "Near the limit" from 80% and "Limit reached" at 100%, and shows when it resets. The monthly limit follows the calendar month. Only Admins and Owners can raise a usage limit above the workspace default.

A Pulsar can also finish on its own and show Done: when it reports the goal reached at two check-ins in a row (if you turned that on), on the finish date you set, or when someone selects Stop for good.

Notifications

By default, a Pulsar notifies the person who created it plus every Admin and Owner. To choose specific people instead, clear The person who created it, plus every Admin and Owner under Who should hear from it? on the Settings tab and select them.

Each person chooses how they hear about Pulsars in Personal Settings > Notifications. For each event, choose Immediately, Daily digest, or Off, and the channel, Email or Teams:

Event Default
New proposals Immediately
Questions Immediately
Proposals about to expire (12 hours before) Immediately
Goal reached / finished Daily digest
Stopped (halted) Immediately

Under Timing, you can turn on quiet hours, which hold notifications overnight and send them when quiet hours end, and set the daily digest time and your time zone. Every notification links straight to the batch, question, or Pulsar. Emails name the proposal and its target but never include tool arguments. Teams options are available once your workspace has connected Teams and your account is linked.

Pulsars on the Home Dashboard

Home has a Pulsars section above Playbooks:

  • Four tiles: Active, Needs you (with the number of proposals and questions), Halted, and Total Pulsars (with drafts and paused). Select Needs you or Halted to open the Pulsars list filtered to those Pulsars.
  • Pulsar activity · 7 days: check-ins per day, finished and failed; proposals by outcome (Approved, Auto-approved, Denied, Expired, and Waiting for review); and Actions carried out, with how many failed

Trial check-ins aren't counted. A workspace with no Pulsars shows a single line with a New Pulsar link instead.

Pulsars in Microsoft Teams

Once your workspace has connected Microsoft Teams, Pulsars work there too. You can approve, deny, and answer from cards in Teams, review Pulsars in a personal tab, and ask Orion about them in chat. On a Pulsar's Settings tab, Also post to Teams channels sends its notifications to channels you choose. Decisions made in Teams are labeled "in Teams" in the web app. See Using Orion in Microsoft Teams.

Who Can Do What

Action Who
Create, edit, start, pause, resume, stop, duplicate, delete, and check in now Any member
Deny proposals, answer questions, and edit or delete what a Pulsar has learned Any member
Approve, approve with changes, and approve all remaining Admin, Owner
Choose Act on approved kinds and the tools it may use without asking Admin, Owner
Raise a usage limit above the workspace default Admin, Owner

Controls a Member can't use stay visible but disabled, with the reason, such as "Only Admins and Owners can approve changes. You can still deny with a reason."

If you are still having trouble, we're here to help! Submit a ticket here for assistance, and don't forget to check our status page to ensure there are no outages in your area.

Comments

0 comments

Please sign in to leave a comment.