Put Bearly on a schedule

Today we're introducing Routines: prompts that Bearly runs for you on a schedule.

Describe one in plain language — "AI news brief, weekday mornings" — and Bearly drafts the name, prompt, and schedule. Each routine gets its own chat. Every run lands there marked unread, and can also arrive as an email or a short text.

Routines run in the desktop app, on your machine. Your routines and their results stay end-to-end encrypted like the rest of your chats, and runs happen on a device you own, not on a server you can't see.

A watcher with a bar

The obvious routine is a morning brief, and Routines do that well. The more interesting ones watch for something.

Say you ship developer tools and want to know when a competitor actually ships — not their marketing calendar, their releases. The routine's prompt: every weekday at 9, read the competitor's changelog and pricing page, check their X account, and compare against previous runs. Two settings change its character:

• Runs build on previous runs gives each run the chat's earlier runs as context, so "what changed since yesterday" is a real question, not a guess.
• Only when sets the bar for interrupting you — here, "a competitor shipped or repriced something."

Runs that don't clear the bar still execute and still land in the chat; they just don't ring any bells — no unread badge, no email. The run history records the reasoning either way, and every 7th quiet run in a row checks in so you know the routine is alive. When the bar can't be checked, Bearly delivers rather than staying quiet on a guess.

What a run can reach

A scheduled run isn't a stripped-down Bearly. It searches the web, searches and reads X, reads and searches the files in its chat — PDFs, saved transcripts, past run output — and writes files of its own. Plugins extend that per routine:

• Connectors, each set to all tools or read-only. A morning routine can read your calendar and inbox without being able to touch either.
• Skills, including their scripts. Flipping on Allow scripts for a skill lets its code run unattended on your machine — with your Bearly token, only for that skill, only after you consented in the editor.
• Deep Research, pre-authorized at the Standard or Ultra tier. Each run does full paid research and emails you the PDF report; the editor shows the per-run price when you pick the tier.

That combination covers more ground than a scheduler usually gets to:

• A researcher drops each week's interview transcripts into the routine's chat as they happen. Sunday evening, the routine re-reads them all and rewrites the running themes memo — what's new, what repeated, which quotes carry.
• An analyst's skill includes a script that reads the warehouse export their BI tool drops on disk every night. The Monday routine runs it, compares the numbers against the forecast saved in the chat, and emails the delta before standup.
• A founder runs Deep Research every Monday at 8 on the week's funding rounds in their space. The report is a PDF in their inbox before the first meeting.
• A weekday 7:30 routine reads a read-only calendar and inbox, cross-checks attendees against the news, and texts the three things worth knowing — a short AI summary to your verified number, capped at three texts a day. A text leaves end-to-end encryption, and the editor says so right where you turn it on.

Unattended, on purpose

A routine runs while you're not watching, so it's deliberately not allowed to act like you. Scheduled runs never see your screen, drive your browser, or take over your computer. Code on your machine is opt-in, per routine: Allow scripts consents one skill's scripts, and turning on unattended local access — past an explicit warning — lets runs use the shell and your files. By default, a routine gets neither.

When a run finds something that needs hands, the handoff is short: its chat is a normal Bearly chat. Open it, and everything interactive is back — take over the browser, use the computer, run local commands — with the routine's full history as context. The routine flags the two invoices that don't reconcile; you sit down and finish it together.

Failure is loud by design. A routine that fails twice in a row pauses itself and tells you — notification and email — instead of quietly skipping mornings. The run history shows every run, every skip, and why. And because routines run on one installed device while the desktop app is open, a closed laptop doesn't cause a pileup: on reopen, the routine runs once for the most recent missed slot and records the older ones as skipped.

Available now

Routines are available today. Create them anywhere — web, mobile, desktop — and install them on the desktop machine that should run them. You can keep up to 10 active routines, at least 30 minutes apart. Find them in the sidebar under Routines or type /routine in any chat. Or skip the page entirely: when a chat is already doing the work, say "do this every Friday at 4" and Bearly turns it into a routine.