Run prompts on a schedule with Routines

A routine is a prompt Bearly runs for you on a schedule — a weekday news brief, a Friday roundup of mentions, a month-end recap. Each routine gets its own chat. Every run lands there marked unread, and you can also have the result emailed or texted to you.

You can create and edit routines anywhere, but they run in the Bearly desktop app, on the one device you install them on, while the app is open.

Create a routine

1. Open Routines

   Select Routines in the sidebar, or type /routine in any chat. You can also ask in a conversation: when a chat is already doing the task, “do this every weekday at 8” turns it into a routine, with the same options as the editor.

2. Describe it

   Type what you want and when into the box — for example, “AI news brief, weekday mornings” — and select Draft routine. Bearly turns your description into a draft. Start from scratch opens a blank editor instead.

3. Review the draft

   Check the Name, Instructions, Model, and Schedule. Write the instructions so the first run has everything it needs, then choose the model you want Bearly to use for every run. The model list follows your plan and your organization’s model policy. Schedules can be daily, weekdays, weekends, chosen days of the week, or a day of the month, at an exact time in your timezone. The editor shows the next few run times so you can confirm it reads your intent correctly.

4. Select Create routine

   The routine gets its own chat, named after it. If you created it on desktop, it starts running there; if you created it on web or mobile, install it from the desktop app when you’re at that machine.

You can keep up to 10 active routines, and runs must be at least 30 minutes apart.

Give runs tools with plugins

In the editor, Tools → Tools and skills controls what each run can use beyond web search, X search, and the files in the routine’s chat:

• Connectors — let runs read your connected apps, like a calendar, or a server you added yourself. Each connector has a Tools setting: All or Read-only. Google Workspace is not available in Routines. A chat approval does not grant unattended Routine access. Connectors sign in per device, so if one isn’t connected on the device that runs the routine, the run says so instead of failing silently.
• Skills — give runs your installed skills. A skill with scripts also has an Allow scripts switch; turning it on lets that skill’s scripts run unattended on your machine. Leave it off unless you trust the skill to run without you watching.
• Deep Research — each run does full paid research and emails you the PDF report. It’s billed per run at the Standard or Ultra tier; the editor shows each tier’s current per-run price, since pricing changes over time. Do the math before you put research on a daily schedule.

Choose the routine’s permission mode

Routines start in Standard, which uses the unattended permissions you have already enabled for that routine. Full access skips additional action approvals for its selected tools. It does not turn on a tool, connect an account, or expand what Routines can do. Your chat’s permission mode does not carry over.

Full access has limited availability and requires a supported account and desktop app version.

1. Create the routine, or select Save after editing it.
2. On the desktop device shown in Run location, open the routine and find Tools → Permission mode.
3. Select Enable full access, review the selected tools and warning, then select Enable full access in the dialog. Choose Cancel to keep the current mode.

The routine shows Full access in its editor and in the list. Enabling it stops any current run and leaves the schedule unchanged. For a managed organization account, an organization administrator must allow Full access first. The routine’s selected tools, connector accounts and tool restrictions still apply, including a connector’s Read-only setting.

Unlike a chat, a routine keeps this permission across app restarts and time away, so scheduled runs can work unattended. Bearly checks authorization for every run. The desktop app must still be open, and its required tools and connections must be available.

Saving changes to the instructions, schedule, model, delivery settings or selected tools stops the current run and requires a new review. Changing a selected connector’s sign-in or an installed skill can also require review. The routine shows Full access needs review and cannot continue in Full access until you select Review full access and approve the current setup on its running device. If its schedule has paused, select Resume after reviewing. Unsaved edits leave the saved routine unchanged. Copies and routines moved to another device start in Standard.

Select Use Standard to turn off Full access and stop the current run. Existing per-tool unattended permissions remain, including local or browser access you enabled separately. Standard routines can still act through those tools without asking; turn off the corresponding tool settings to remove that access. Stop run ends just the current run and keeps its schedule and permission mode. A change made from another device takes effect when it syncs to the running device. Web and mobile can manage these settings, but cannot enable Full access or run the routine.

If Permission mode cannot load, check your connection and select Retry. Follow any message to update Bearly or enable an available tool. If your organization no longer allows Full access, choose Use Standard or ask an organization administrator to restore eligibility. Restored eligibility still requires a new review; it does not resume the routine automatically.

Full access does not add screen or computer control, Google Workspace, or Chrome DevTools to Routines. Browser access stays limited to the routine’s own tabs and to the sites and switches you set under Let a routine use a browser.

Allow commands and local file access

On the desktop app, Tools → Unattended local access lets a routine use Local Exec and selected skill scripts without waiting for approval. Bearly shows a detailed warning and does not turn the setting on until you select Allow unattended access. If you ask Bearly to enable Local access while setting up a routine in chat, an approval card with the same risks appears before the routine is saved; select Allow to continue or Deny to stop the save without enabling it. Asking Bearly in chat to change the instructions of a routine with Local access shows the card again, because the instructions are what you approved.

The routine can change or delete data

With unattended local access on, the AI can run shell commands and skill scripts and can read, upload, change, overwrite, or permanently delete files available to your computer account. It acts without asking before each command. AI mistakes or malicious instructions hidden in websites, files, email, or connector data could expose secrets or destroy data.

Leave this off unless the routine needs local commands. Keep current backups, give the routine the smallest practical working folder, and test the saved routine with Run now before relying on its schedule. You can turn the setting off at any time; later runs will stop receiving Local Exec.

This setting does not override your computer’s permissions or your organization’s Bearly policy, and it never lets a scheduled run see your screen or control your computer. Browser access is a separate setting, described next.

Build on previous runs under Results and delivery is on by default for new routines. Each run sees the chat’s earlier runs — useful for “tell me what changed since yesterday” or “don’t recommend anything you’ve already suggested.” Turn it off when every run should start clean. Existing routines keep their saved setting.

Let a routine use a browser

On the desktop app, Tools → Browser access lets scheduled runs use Bearly Browser on the sites you list, while you’re away. Nothing it can do there needs your approval at the time, so the list and the switches below are the limits, with or without Full access. Routines can’t use your Chrome.

• Sites — the exact sites the routine can open, like example.com. www. is optional, but every other subdomain is its own site: add pay.example.com or a sign-in page on another domain separately. Sites named in the instructions appear under From the instructions so you can add them in one click. The routine can’t open, read, or act on any other site, and a page it opens can’t send it to one: the desktop app stops the request before it goes out. The routine says where it stopped in Recent runs.
• Click, type, and submit — off by default. When off, the routine only reads pages and follows links. Turn it on when the routine has to click buttons, type, choose options, or submit a form.
• Download files — off by default, and available only with Click, type, and submit on. Lets the routine save files from these sites into its chat.

When you save, a warning lists everything the routine can do and what’s new since you last approved it, and nothing is saved until you select Allow browser access and save. Removing a site or turning a switch off saves without asking. If you ask Bearly in chat to set up browser access, an approval card with the same details appears before the routine is saved; select Allow to continue or Deny to save nothing. The card appears again whenever the access grows, or when Bearly changes the instructions of a routine that has browser access.

The routine uses the site sign-ins in Bearly Browser on the device that runs it, including cookies you imported from Chrome or Edge. The Browser panel stays closed while the routine works, and the routine’s tabs close when the run ends. Enable Browser and Allow chats to use Browser must be on for that device, and the desktop app must be up to date.

The routine acts as you on the sites you allow

With Click, type, and submit on, a run can send messages, make purchases, and change or delete account data on those sites without asking. Instructions hidden in a web page, email, or document can trick it into actions your instructions didn’t ask for, and sites treat its actions as yours.

Allow only the sites the routine needs, leave Click, type, and submit off unless it has to change something, and test with Run now before relying on the schedule. Sign in ahead of time: a run never clicks or types into password or payment fields, and can’t upload files. No run accepts an invalid HTTPS certificate; the routine closes that tab and reports the problem.

If the instructions or the browser access change on a device that can’t check them, such as an older version of Bearly, later runs go without a browser until you open the routine and save it to approve them again. Routines that had browser access before sites existed also run without a browser until you pick their sites.

If Bearly Browser isn’t available when a run starts, the run goes ahead without it and says so in Recent runs. That happens when your organization doesn’t allow it, when Browser is off on that device, or when the desktop app needs an update. A Full access run also needs current authorization for its remaining tools; if its permissions need review, it stops instead. You can turn browser access off at any time; later runs won’t use a browser.

Control when you hear about it

Every run lands in the routine’s chat either way. The Notification filter field sets the bar for actually notifying you — for example, “a competitor ships something new”. When a run doesn’t meet the bar, there’s no unread badge, email, or text; the result just sits quietly in the chat, and Recent runs notes why. So that silence never reads as a dead routine, it checks in anyway every 7th quiet run in a row — and when the bar can’t be checked, it delivers to be safe.

Two delivery options sit below it. They tell the routine’s chat to send the finished result before the run ends, so the send appears in the chat with the rest of the run:

• Email each result — sends each delivered result to your account email. Routines only email you, never other addresses.
• Text each result — texts a short AI summary of each delivered result to your verified number. If texting is not ready, turning this on opens Settings → Phone & SMS. Add or re-verify your number there; after verification, the routine’s text option turns on. Texts are capped at 3 a day, and a text leaves end-to-end encryption — the switch says so where you turn it on.

If a send fails, open the routine’s chat and read the failed Send Email or Send SMS step. When SMS needs setup, the chat also shows an Open Phone & SMS button. The run result stays in the chat, and you can ask Bearly to send it again after fixing the account or phone setting. Check whether the first message arrived before retrying so you do not create a duplicate.

Where routines run

Routines run on one device: the Run location row in the editor shows which. Install on this device claims a routine created elsewhere; Move to this device takes one over from another machine. A routine that isn’t installed anywhere doesn’t run.

The desktop app has to be open. If it was closed or asleep over a scheduled time, the routine catches up when the app is back: it runs once for the most recent missed slot, and older missed slots are recorded as skipped in Recent runs rather than run late in a burst.

Run now in the editor runs the last saved version immediately on the current device — useful to test a new prompt without waiting for the schedule. A Full access routine must run on its installed desktop device.

When a routine fails

Recent runs in the editor shows each run’s outcome, including skips and why. If a routine fails twice in a row it pauses itself and shows a Bearly notification. Fix the cause — often a connector that needs reconnecting on that device — then select Resume. Individual runs are also stopped if they take longer than about 25 minutes.

A run that uses Bearly Browser stops if you reset Browser tabs, delete Browser data, or turn off Browser while it works; Recent runs says so. If you take over one of the routine’s Browser tabs, the run can’t use that tab and reports what it finished.

If you delete a routine’s chat, its next run creates a new chat with the routine’s name. Standard routines continue there; a Full access routine needs a new permission review before continuing in the replacement chat. The deleted chat and its contents are not restored. If Bearly cannot create the replacement, the routine pauses and Recent runs shows that the chat was missing.

Pausing a routine stops its schedule and current run but keeps its settings and history. Stop run stops current work without pausing future runs. Deleting a routine stops it; its chat and past runs stay in your history. Changes made on another device take effect on the running device when they sync. None of these controls undo actions already completed in another system.

What scheduled runs can and can’t do

A scheduled run works unattended, so it’s deliberately more limited than you working in a chat. It can search the web and X, read and search the files in its chat (including PDFs and saved transcripts), write files, use the connectors and skills you selected, and run paid Deep Research if you enabled it. By default, it does not run general commands on your machine; skill scripts run only for skills where you turned on Allow scripts. Turning on Allow unattended local tools also gives the routine Local Exec and permits selected skill scripts without individual approval.

Scheduled runs never see your screen or control your computer. They use a browser only when you turn on Browser access, and then only Bearly Browser, on the sites you list.

For anything that needs your hands or eyes, open the routine’s chat and continue the conversation there. It’s a normal Bearly chat with everything you have anywhere else, and the routine’s full run history as context.

Administrators of managed organizations can turn Routines off from Organizations → organization name → Policies → policy name → Code & Automation → Routines. Browser in Routines in the same section controls browser access; a routine also needs Browser allowed.

For organization eligibility and the difference from interactive chats, see Choose how Bearly uses external tools.