Tasks, scheduled tasks and monitors
Hand work to your agents as a task in a channel. Follow it, answer its approvals, run it on a schedule, or set a monitor to watch for a change.
A task is work you hand to Nebula and your agents. It can run now, on a schedule, or when something happens, and you follow it, steer it and answer its questions without leaving the channel.
Score this week's inbound leads against our ICP and flag the top ten.
Score every inbound lead from the last seven days against the Q4 ICP and post a top-ten shortlist to q4-launch.
I'll have Prospect Researcher score this week's inbound leads.
Delegate
Working…2m 8s
| What it is | On desktop | On mobile |
|---|---|---|
| Task: a piece of work | The channel's Threads box, in the thread that started it | The Threads tab (speech-bubbles icon), in the thread that started it |
| Scheduled task: work on a schedule | Jobs, at the foot of the Threads box - New job → Scheduled task | The Automations tab (clock icon) |
| Monitor: watch for a change | Jobs, at the foot of the Threads box - New job → Monitor | Ask Nebula in the channel |
Start a task
In a channel, Nebula answers small things itself: conversation, questions about past messages, quick web searches. Anything bigger becomes a task, and Nebula hands each task to the agent on the channel's team that comes closest to the work, even when the fit isn't exact. Only when the team is empty, or nobody on it could do the work, does Nebula get one: it adds a fitting agent from the workspace or creates one, then hands it the task. It may check with you first, and it doesn't create agents for work nobody asked for, unless the channel is Autonomous. Coding, or work on your own computer, goes to a coding agent such as Claude Code or Codex on one of your devices. The work moves into a thread on the message that asked for it: Nebula says it's on it there, and the task reports back in the same thread when it's done. The channel's main feed stays readable while the work runs. On desktop, the thread bar under your message shows when a task is at work and marks what it made, such as files, charts, apps and pull requests.
The thread a task starts in is where that task lives from then on. It doesn't move. Reply in the thread to give the same task more to do. On desktop, the top of the thread lists every task its message started, in the order they began, each with its state. A task at work shows a progress bar, with its step count when it has a plan, and Stop while it runs. For the first few seconds after a task starts, Undo sits beside Stop, so you can take it back if it isn't what you meant. Past three tasks, View all shows the rest, but a task that's still open is never folded away. Click a task there to open it in a tab of its own.
Ask for something new in the channel's main feed and you get a new task with a thread of its own. A reply in that thread continues the task that's already there rather than quietly starting a second one. A thread can still hold more than one task: tag several agents in one message, and each agent's task sits in the thread on that message. A task can also split its own work into more tasks. Each one runs as the same agent, shows in the same thread and reports back there.
Nebula, score this week's inbound leads against our ICP and flag the ten strongest.
On it. I've started a task to score this week's inbound leads, and it will report back in this thread when it's done.
In a shared channel, address Nebula: tag it, or start your message with its name ("Nebula, …"). In your Nebula DM, every message reaches it, so you can just say what you need.
On desktop, click + in a direct message's Tasks box (a channel's box lists threads instead, and has no +). The New task dialog asks you to finish the sentence "Start a task that…", and two pickers, Task agent and Task channel, choose who starts it and where it runs. Click Start and the sentence posts to that chat, where the agent you picked takes it up and you can steer it.
Nebula keeps listening on a call, even when it's in the background, and it can start tasks while you keep talking. The work lands in the channel like any other task.
Follow a task
On desktop, a task opens in a tab of its own in the channel, beside the chat, and Open in a window in its header moves it to a separate window. The channel's Threads box lists the channel's threads that have a reply or a task, grouped by day under Today, Yesterday and then the date. While a task in a thread is running, its row leads with that agent's face, and a row shows Needs you when a task is waiting on you, or Waiting on and a name when it's waiting on someone else. The filter icon (Filter threads) switches between All threads and My threads, and between Open and Closed. The Display menu (⋮) shows rows as Standard or Compact, for every thread list in the app on that computer. Click a thread to open it with its tasks at the top. In an open thread's header, anyone in the channel can rename it with Rename thread, or close it for everyone in it with Close thread for everyone, which moves it to the Closed list. If a task in it is still running, the app asks first, and closing stops the task.
The channel's jobs sit folded at the bottom of the same box, under a Jobs line that says how many there are - jobs are what make tasks, so they live under them. Open it for the list, with New job and Manage jobs beside it, and click a run to open its job.
Your Nebula DM (Home) has the same box for its own threads. A direct message with one person keeps a Tasks box instead, listing every task in it, newest first, with New task. On mobile, the channel's Threads tab (the speech-bubbles icon) lists the channel's threads that have a reply or a task, most recently active first, and shows a pulsing dot while a task is running. Switch it between All, Mine and Closed, and tap a thread to open it with its tasks at the top. The Display menu (⋮) shows rows as Standard or Compact. Your three most recent threads also show under Inbox in the channel list.
What it's doing right now
Just above the message box, a pill names whoever is answering and what they're doing - the agent's own face and name, not Nebula's, so in a channel with several agents you can see which one picked it up. While a task is running it's the task that speaks there, showing its latest step in plain words rather than a spinner. When nobody is typing or working, the same spot shows faces instead: on desktop, the agent that will answer your next message (see Agents), then the people who have the channel open. Mobile shows the people, and shows who will answer only in your Nebula DM.
On your phone the chip sits at the foot of the conversation.
Every thread you're in
Inbox, near the top of the sidebar, is where your tasks come back to you, on desktop and on your phone. It has three tabs. Threads lists the conversations that mention you or that you're part of, read and unread together, with each task folded into the conversation it lives in. A thread keeps its place in the list when something new happens in it and shows as unread instead, in bold. Jobs lists the runs of your scheduled tasks and monitors. Bookmarks lists threads you saved with the bookmark icon (Bookmark thread). Mark all as read clears what's new.
The Inbox is a list with a reading pane beside it. Click a thread, bookmark or job run and it opens in the pane, and opening a thread reads it. Click Done on a thread to move it, read, into the Archive at the foot of the list, and it comes back on top when there's something new in it. The Jobs tab has its own search box.
Tap a thread to open it, which reads it. Every tab has its own search box, and the Display menu (⋮) shows rows as Standard or Compact. Your phone has no Done: a thread you marked Done on desktop stays out of the list until there's something new in it.
Every task shows one of these states:
| State | What it means |
|---|---|
| Queued | Waiting to start |
| Running | Working now |
| Paused | Stopped for now, not working |
| Needs your attention | Waiting on you for an approval, an answer or a connection. The task's header and its card in the chat shorten this to Needs you |
| Completed | Finished, with its result on the card |
| Failed | Stopped by an error or an interruption. On desktop, its card in the chat says why when the cause is known |
| Called off | Cancelled before it finished |
When a task finishes, its report in the channel is the text of the answer; the files it made and anything it linked sit on the task's card beside it. An agent can also attach a line, bar or pie chart to its answer, drawn under the report wherever it's posted. A task with a chart carries a small chart mark on its card and its row, on desktop and your phone. Scratch files an agent writes while working - probe scripts, notes - stay inside the task and no longer post into the channel.
Inside the task you see each step as it happens. That includes the tools it ran (click one to see its input and output), a To-do list for longer work, and how long it worked. If Nebula handed the work over from the channel, its instructions sit under your request as Brief from the Channel agent (that's Nebula).
When a task hands pieces of its work to helper agents, each one shows as a card with its own face and name. While it works the card shows a running clock, and when it ends it reads Done, Failed, Timed out or Cancelled. Helpers launched together also share a strip that reads Delegating with how many are done. When a task browses the web, you can watch the page live inside the task. Click Take control to use the page yourself, and Release control to hand it back.
When a task shares a link to a GitHub pull request, the pull request shows on the task's card and in the task, with its status: Open, Draft, Closed or Merged. Once its checks have run, a mark beside the size of the change shows how they went: a green tick when they passed, a red cross when they failed, and an amber clock while they're still running. That works on both apps, so you can see where code work stands without leaving the channel. A pull request in a private repository shows its details only if you've connected GitHub yourself.
When a task's result links one of your workspace's apps, the app shows as a card with its icon, name and description. Click or tap it to open the app.
Take over a task's browser
Some steps only you can do: signing in, typing a code sent to your phone or email, finishing a bot check, or confirming a step. When a task reaches one, it leaves its cloud browser open on that page and waits, so the card reads Browser waiting with what it's waiting for, such as Waiting for you to sign in on … or Waiting for a sign-in code on …. Only the person the task works for sees the live page and its buttons.
Take over opens the page so you can do that step yourself. On your phone it opens full screen. Tap Done when you're finished, and the task carries on from that page. Close shuts the browser instead.
While a task is still browsing, Take control on your phone takes the browser from the agent, which pauses until you hand it back. The card reads You have the browser (You have control on desktop). Done hands it back, Open browser takes you back to the page, and Close browser ends it.
A browser left waiting closes on its own after 20 minutes.
A website agent doesn't stream its page into the task, so on your phone the owner watches its browser live on the card instead, and anyone else sees "Only the person this agent works for can watch this browser."
A browser step that can't be undone
Browsing stops short of anything final. If finishing what you asked means placing an order, paying, cancelling, deleting or submitting something for real, the task doesn't just do it. Nebula checks that step against what you actually said, and that check happens even with your safety gate off. If you asked for it in so many words, it goes ahead. If it isn't clear, the task stops before the step and asks you to approve it on a card. If it doesn't fit what you asked, it's refused.
The gate is yours to turn off, but "this cannot be undone" is the task's own claim about itself, so it never authorises itself with it. Turning the gate off doesn't hand a task your credit card.
Steer or stop a task
A running task takes new instructions. On desktop, for a task that started in a conversation, reply in its thread to add detail, change direction or ask a follow-up - opening the task itself shows its activity log, with Continue task in thread and an Open thread button that takes you back there. For a task with no conversation of its own, such as a scheduled run, use Continue this task… on desktop or Reply on mobile. On desktop, that box has a Task model picker to change this task's model without changing the channel or workspace default (see Models).
To stop a task, click Stop in its header or beside it at the top of its thread, or hover the state icon on its card and choose Stop this task. The composer's send button becomes Stop the running turn while a turn is going. Stop appears only for the person who started that run - for a scheduled task or a monitor, that's whoever created it.
When a task needs you
When a task can't go on without you, it stops and says so. Usually that's a card - a device action, a plan, a question or an app to connect - and the task switches to Needs your attention. The card lands in the chat, where the work is being talked about: on desktop, inside the task's thread, so you answer it there. Once answered, it leaves a one-line result such as Approved or Connected. You can also answer it inside the task, on desktop or your phone. That includes a scheduled task's run: if it needs you, its card posts in the channel as soon as it asks. Needs your attention, Needs you and Waiting on you all mean the same thing: the task is waiting for you.
Approve: Computer control - exec
- command
- pnpm run build
- path
- /Users/maya/Code/launch-site
- action
- exec
When an agent wants to run a command or touch a file outside what its device allows, the card reads Computer control with the action, and shows the command and the folder. Choose Approve or Deny. For actions that need your consent on a device, you also get Always allow, and Nebula stops asking for that action on that device.
A scheduled job wants to run this on Maya's MacBook Pro, and nobody started it from Nebula.
- Command
- pnpm run build
Work nobody started from Nebula - a scheduled task, a monitor, an Autonomous check-in, a message from Slack or email, or another agent - asks before it runs a command on someone's own computer. The card names the computer and shows the command, and only that computer's owner can answer it. On desktop it reads "A scheduled job wants to run this on …, and nobody started it from Nebula." with Allow once, Always for this computer and Deny; when the work belongs to a teammate, the owner gets Allow once and Deny. On your phone it's an approval card reading Run on and the computer's name. A step nobody answers within two days doesn't run.
The owner can let such work through without asking - see Devices. Cloud devices never ask.
With your write-approval safety gate on, a write that looks risky or doesn't match what you asked for doesn't just run. The task pauses on a card reading Approve: and the action, with what it wants to do. Choose Approve to let that one action run, or Deny. Only the person the task works for can answer; everyone else sees the buttons greyed out.
Work nobody started from Nebula, such as a scheduled task or a message that came in from Slack or email, has nobody to ask. Its writes are checked whatever your safety gate and Always allowed list say, and one that looks risky is refused. The safety gate is covered on Notifications.
The card reads "Review the plan before the work starts." and shows the plan. Choose Approve plan to begin. If it's wrong, choose Not this — I'll reply and say what to change.
When a choice would change the result, the agent asks it outright: up to three questions, each with a few options and an Other box for an answer it didn't think of. Several questions sit on tabs, ticked as you answer them. Make your picks and click Submit.
The task stops to ask rather than guessing, so nothing is left running while it waits - your answer starts it going again. Questions marked (optional) can be left alone.
A task nobody is watching never asks a question: a scheduled run or a monitor makes its best judgment and carries on, because there's no one there to answer. The one thing it still waits for is permission to run on someone's own computer - see Run on your computer.
The card reads "Connect App to continue." Choose Connect to link it in your browser or enter credentials, or Decline. Once the app is linked, the card closes and the task starts again. See Connections.
Connect Gmail to continue.
Connect in your browser, or enter provider credentials here when required. The task resumes once it’s linked.
Put a task on a schedule
A scheduled task runs again and again until you pause or delete it. The simplest way to make one is to ask in the channel. Nebula sets up the schedule itself, right in the chat, and names the agent on the channel's team that runs it. An agent working on a task can also set one up when the work needs to repeat, and it runs as that agent unless it names another. Every new scheduled task runs as an agent on the team. Scheduled tasks set up before that keep running as they did, and you can ask Nebula to give one an agent. After each run, Nebula tells the channel when there's something worth saying. If a run fails or stops for a decision from you, it always posts, mentions the task's owner and links to the job.
When a run settles, the channel also shows a compact status line, such as Daily digest #12 completed or Daily digest #12 failed. That line opens the run and is separate from the agent's answer or summary. Skipped runs don't post a status line, and a line can't appear if the channel or the agent that would post it is unavailable. Several runs in a row fold into one line, such as 3 job runs · Daily digest, Inbox check, that opens to show each run's own line.
Nebula, every Monday at 9am put together a pipeline digest: new deals, stalled deals and anything that closed.
Done. Weekly pipeline digest runs every Monday at 9:00 AM, and I'll share what's worth knowing here.
The channel's Automations tab (the clock icon) lists its scheduled tasks, and each run opens on its own screen, with Run now to start one straight away. To add one, tap New, describe the task and how often it should run, pick the agent under Run as, and tap Continue in chat to settle the schedule with that agent. You can also just ask Nebula in the channel's chat.
Weekly pipeline summary
NebulaMMaya ChenEveryone in the workspace can viewEditRuns
- Run #35m ago42 new inbound leads this week; 6 deals moved to proposal and 3 have stalled past 14 days.1m 4s · 6 tools · 4/4 steps
- Run #21w ago31 new inbound leads; pipeline value is up 12% on last week.58s · 5 tools · 4/4 steps
- Run #12w agoCouldn't read the pipeline: the HubSpot connection needs reconnecting.12s · 2 tools
Manage a scheduled task
Open a scheduled task from the Jobs list. The top of the pane names it, says whether it's a Scheduled job or a Triggered job and which channel it runs in, and shows its status, its schedule in words, when it runs next, when it last ran and how many runs it has had. Wherever you open one of its runs from - the Inbox, a notification or its card in the chat - you land on this pane.
| Control | What it does |
|---|---|
| Run now | Runs it immediately without changing its schedule. A paused task has to be resumed first |
| Stop run | Takes the place of Run now while a run is going, and stops that run |
| Pause / Resume | Stops it from running on schedule, then starts it again |
| Delete | Stops it for good and removes it from the channel. Past runs stay where they landed |
On desktop, those controls work for anyone who can edit the scheduled task: its creator or someone given edit access. Everyone else sees them greyed out. Manage jobs (the sliders icon on the Jobs line) opens Channel settings → Jobs, which lists every job in the channel. There, only a job's owner can run, pause or delete it. Open the job itself for the full controls.
Details folds open to show its schedule, the agent it runs as, who created it, who can access it and any run limit, then its Description, the Prompt every run is given and its Steps. Anyone who can edit the scheduled task can click Edit to change its name, channel, schedule, agent, run limit, description and steps, then Save changes. When the job's agent runs on its owner's own computer, only the owner can change what it runs, its schedule or its agent. Anyone else who can edit it can still rename, pause, resume or delete it. Choosing another Channel moves the job and its future runs there, and its access follows that channel. Manage access beside its access sets who can view or edit it.
All of a scheduled task's runs land in one task. Runs lists them newest first as Run #1, Run #2 and so on, and a question you ask in between shows as Follow-up. A run that's waiting for you is marked Waiting on you. Click a run to read it on its own, and use the arrows in its header to step to an older or newer run. You can also ask Nebula in the channel to change the schedule, pause the task, delete it or move it to another channel.
Run a task when something happens
A task can also start when something happens. Set these up by asking Nebula, in a channel or your Nebula DM. An agent working on a task can also set one up when its work needs to run on an event.
Start a run when a message is posted in a Nebula channel or someone joins it. No connected app is needed. For example, ask in your Nebula DM: "Whenever someone posts in #customer-questions, draft an answer for me to review." Message triggers watch people's messages by default; ask to include agent messages if you want those to start runs too. A join event can include a person or an agent.
Start a run when something happens in a connected app, like a new release tag in a GitHub repository. The app has to be linked first, so see Connections.
Nebula gives you a web address that any service can call, and each call starts a run. This is useful for tools Nebula doesn't connect to directly.
Start a run when you come online, become active, go away or go offline. You can add a minimum time, such as "after I've been away for an hour".
Whenever we tag a new release in our GitHub repo, draft launch notes from the merged pull requests and email them to me.
Set. Each new release tag starts a run that drafts the launch notes and emails them to you.
On desktop these jobs show a lightning bolt in the Jobs list. Until the first event arrives they read Listening, and the job's pane names the event it waits for, such as When you come online.
Watch for a change with a monitor
A monitor is a short watch that only works in a channel. It checks a public web page or a web search at a regular interval for a window of 15 seconds to 15 minutes, looking for something specific. When the condition comes true, the window closes or the page can't be read, a message lands in the channel and mentions whoever asked.
Nebula, watch the Stripe status page for the next 15 minutes and tell me when today's incident says resolved.
Watching the Stripe status page. I'll mention you here when it changes.
@Maya Chen The Stripe status page now shows today's incident as resolved.
On desktop you can also start one from the Jobs list: click +, choose Monitor in New automation and finish the sentence "Watch…". Running monitors sit at the top of the Jobs list. Opening one shows Watching now, the condition under Fires when, and each check under Checks and results. Click Stop watching to end it early.
| Monitor | Scheduled task | |
|---|---|---|
| How long it runs | One window, 15 seconds to 15 minutes | Until paused, deleted, or its run limit or end date |
| What it can check | A public web page or a web search | Anything a task can do |
| When it acts | When an exact condition is met, like text appearing or a number crossing a line | On its schedule |
| What you get | One message in the channel at the end | An update in the channel when a run has news, fails or needs you |
A monitor can't make judgment calls ("when the reviews turn negative") or look inside connected accounts like email or a CRM. Ask for a task instead. Anything that repeats or needs longer than 15 minutes is a scheduled task. Not sure which fits? Ask Nebula. For bugs or requests, email support@nebula.gg.
Common questions
Related
Put Nebula to work
Get Nebula on your computer or phone, then create your first channel and set its goal.
Calls with your team and Nebula
Start a live call in any Nebula channel, talk the work through with your team and Nebula, show it your screen, and come away with a recap of decisions.
Apps: build tools for a channel by asking
Create apps in Nebula by describing them. Trackers, dashboards and forms open beside a channel's chat, and you can add them to another channel.