Omniscio

Tasks (markdown outliner, in development)

Tasks is the app's only task list: a keyboard-driven outliner where every task is a line of markdown, grouped into projects and shown one project at a time (or, with Show all projects in one scroll on in Customize, as one All Projects list). It is still in development and off until you turn it on in Settings → Features. Once on, it adds a Today plate, AI planning helpers and agents you can hand a task to.

On this page

What it is

Tasks is the app's task outliner, surfaced as a virtual project in the sidebar labelled Tasks: a keyboard-driven tree of to-do rows where each row's text is real markdown, rendered in view mode and swapped to a raw-source editor on click. Single-click (or tap) a row to highlight it; double-click it or press Space to edit. Enter adds a new task directly below (Shift+Enter adds one above), and on a phone a tap only highlights, so you edit through the focused row's ✏️ Edit button.

Rows nest with Tab / Shift+Tab — Tab indents only the focused row, leaving its sub-items at their depth, re-homed beside it under the new parent. They reorder by dragging, the Move up / Move down rows in a row's ⋯ menu, or Alt+Shift+↑/↓, and nest by dragging a row onto the middle of another. Hand-reorder only applies in the manual sort with no filter: under any other sort or a filter the on-screen order is a view overlay, so those controls do nothing. Ctrl/Cmd+D duplicates a row with its open sub-tasks, notes and links; Ctrl/Cmd+Delete archives it. Rows carry due / snooze / context / importance / estimated-time / energy / tag chips.

A row's Task details pane (the sliders icon, or Cmd/Ctrl+;) edits its due date, repeat, snooze-until, context, importance, scheduled start, estimated time, energy level, context tag and tags, each auto-saving as you leave it — there is no Save button. An Urgency dial appears beside Importance when Urgency is turned on in Tasks → Customize (off by default).

Completing a parent marks its undone descendants done. Archiving a row files its subtree into Archived, restorable — except a sub-task snoozed for later, which is never swept when you clear its parent: clearing a parent spares each parked sub-task and lifts it to the top level of its project, so it still returns on schedule.

Tasks are organized into named projects, each a workspace with shared master context (a markdown note pinned above its tasks). Every task belongs to exactly one project; the default Inbox holds anything unfiled. By default Tasks shows one project at a time; turn on "Show all projects in one scroll" in Customize for one giant list instead, with projects as collapsible top-level nodes. Find (/), bulk-select (Ctrl/Cmd+A) and repeating tasks ("every monday") work everywhere, and you can pin projects as extra columns (drag onto the main panel, the project menu's Pin as column, or Shift+Enter) — see The giant default view and Pinned columns.

Tasks persist in a local database and mirror to one plain-markdown tasks-v2.md file per account, so other tools and local AI agents can read the tree; the mirror is written only while Tasks is on and Database Encryption is off (with encryption on, the app deletes the plaintext copy on purpose — see part 10). Tasks is a separate feature from the original Tasks (v1) outliner, since removed.

It opens instantly. Tasks paints what it loaded last time on open and refreshes behind you, so the skeleton only appears on the first-ever load. On desktop the panel is also kept warm (mounted but hidden) for an instant switch back. Opt out at Settings → Performance → "Keep Tasks loaded for instant switching" (tasksV2KeepWarmEnabled, default on, desktop only) — the only switch, and never applied on a phone, where Tasks opens as an overlay.

Where to find it

In development and off by default — turn it on at Settings → Features. Once on, Tasks appears as a project in the sidebar. While it is off, opening Tasks (for example from the Productivity hub) shows a "Tasks is turned off" screen with an Open Lab button that jumps straight to the Tasks switch.

Switching Tasks off never deletes your tasks. It only hides the feature: your active tasks stay in the database and come back, unchanged, the moment you switch Tasks on again. What does pause is the background work — due, snooze and follow-up reminders (including the email, text and phone channels), session-end catches and the plaintext markdown copy stop at their next run and start again once Tasks is back on, so nothing is reminded, caught or updated for you while it is off — unless you turned on "Keep trickles running while Tasks is hidden" on the Tasks settings page (in the Tasks display group — search Settings for "trickles"), which keeps trickle releases sending tasks to your inbox and filing them while the feature stays hidden. A Do it for me draft is not paused, though: a draft still being written is stopped and marked failed, and it does not restart — delegate the task again once Tasks is back on. Switching off Tasks delegation (AI does the task) in Settings, Lab, or signing out, stops a running draft the same way. Housekeeping keeps going: tasks you had already deleted are still cleared out on the usual 30-day schedule.

How it behaves

The view at a glance — progressive disclosure

The default view is deliberately quiet: the project rail on the left, a one-line header titled with the open project's name, and that project's tasks. With "Show all projects in one scroll" on, the header reads "All Projects" and every project appears under a collapsible heading with its own "+" to add a task to it. Everything else opens on demand from the header:

The header carries — title left, the rest right-aligned:

  • the title is itself the "Show ▾" dropdown (desktop) — pick a single project, or All Projects once "Show all projects in one scroll" is on; on mobile the title is the rail switcher.
  • a "Due" badge, shown only when something is genuinely due soon (overdue / today / this week) and counted by the same rule as the Due list, AI-inferred deadlines included, re-checked each minute. Hovering it shows a tooltip with the count ("3 due soon"), like the other header icons. The badge and the list read the same tasks (every project's, through one count), so the number on the badge is always the number of rows the list holds; when the list is open and nothing is left due, it says "Nothing overdue or due this week." instead of going blank. Click it for the cross-project Due this week list, whose Overdue section offers Reschedule all (N) (one new date, one Undo) and Spread them out (5 a day over the next workdays, skipping weekend days, one Undo).
  • a Filter funnel — sort, importance threshold, due date and tag toggles, with a dot when one is active. While a filter hides rows, each project heading's count reads an honest "n of m", and a fully hidden project shows "N hidden by filter — Clear filter" rather than a silent blank.
  • a Find button (🔍) (the same bar as / and Ctrl/Cmd+F).
  • a "⋯ More" menu, organised like the session ⋯ menu: five sections that are all closed each time the menu opens, of which one opens at a time (opening another closes the first), with Customize always at the foot outside them. A section with nothing to offer on the current screen is not shown.
    • This project (only with one project open): Launch agent…, Snooze all…, Priority… and Master context.
    • Plan: daily check-in, Plan my day, Label tasks by urgent & important and About you.
    • Import & export: send-to-inbox, export JSON, Import tasks… and Import notes from Obsidian….
    • View: Pop out (desktop app only), Archived and Recently deleted.
    • Help: Keyboard shortcuts (desktop only, the same cheat-sheet as ?) and the guided tour.
    • Customize opens the whole Tasks settings surface in the content area as four cards — Set up Tasks (the AI, Caught, Memory and Mission Control switches, the live AI status line and Turn on all), How the list looks, How tasks behave and AI features — plus "More task settings…". While it is open the list and every control that acts on it step aside, and closing it returns you to the same view, scroll position and selection. The left rail keeps only Archived (with its count) at its foot. Each Customize toggle also has a searchable row in Settings under a Tasks section with no sidebar row.

All the date surfaces share one rule. "Overdue" means the deadline itself has passed — a task due at 9am today reads overdue from 9am on, a task due at 6pm today reads due today until 6pm, and a deadline on an earlier calendar day is always overdue; "this week" is the calendar week, ending at local midnight before next week's first day as the Calendar week-start day sets it (Monday by default). The Due list, due chips, Due badge, Today's DUE label, the All Projects today band and the toolbar filter all ask that question the same way; Today's label reads OVERDUE, DUE TODAY or DUE <WEEKDAY> inside the current week and DUE <MONTH DAY> beyond it, and the evening close-out grades its risks by that same question through dueMissedAtIso.

When a single project is in view, a "+" button puts the cursor in that project's Add a task box (the same box the All Projects view's project "+" uses), on desktop and phone. The whole-project actions live in ⋯ More → This project on every screen — Launch agent (one agent on the whole project, given only the first 40 open tasks in list order, sub-tasks included), Snooze all and Priority (its picker opens from the ⋯ button) — hidden in All Projects and Today, the last two disabled on an empty project. Every popover stays inside the window and closes on Esc or an outside click.

On each task row the actions are two primary buttons (Task details, Launch agent) plus a "⋯" menu: Add subtask, Copy, Cut / Paste here (shown only when something is cut), Context, Break it down, Move to project, and in the manual sort Move up / Move down. On mobile a leading ✏️ Edit button appears too, and Attach a session lives on the right-click menu and ⋯ → Context. On desktop the cluster reveals on hover (or keyboard focus), flows inline with the text so a long row wraps it, hides while you edit, and goes quiet the moment you navigate by keyboard. On a phone it stays visible on the tapped row with 44px tap targets. Right-clicking a row (desktop) opens the same action set as a cursor-positioned menu, including the Style row (Star / Bold / Highlight), Insert link, Get link title and Finish. A clickable 📎 count chip peeks at a row's attachments. See Cut & paste a task.

On a phone, Tasks' small controls get a 44px tap area, and the desktop look is unchanged. The "Break this down?" hint, the attachment "Copy details" link, project headings, the list title, the context peek rows and the empty "Add shared context" prompt grow to 44px; tag filter chips already did. Controls drawn with the shared Button already carry the floor. The Tasks Shortcuts settings controls (remove key, Default, Confirm, Cancel) and the display-only Drafting chip carry pointer anchors, so the assistant can point at them.

Milestone nudges

As you use Tasks, brief toasts mark natural progression points — your 5th task created (suggesting projects), your first overdue task (introducing the Today view), your 10th task (keyboard shortcuts) and your first project (daily planning). Two more name one of your own tasks: a big task (long, important or dated enough) that has sat 2+ days with no open sub-steps gets "…looks like a big one. Break it into smaller steps?" with a Break it down button, and a task 3+ days overdue gets "…has been overdue for a few days. Snooze it until a day that works?" with a Snooze it button. Long names are shortened to 40 characters. These two never count welcome, imported or another account's tasks, nor finished, archived or snoozed ones, and the breakdown tip stays quiet while Break it down or Tasks AI is off.

Only one tip shows at a time, the first-overdue Today tip going first, and a waiting task tip whose task has since gone waits for the next matching task. Each count is the signed-in account's own work, and the four starter tasks never count. Each toast has an action button and a "Don't show these" opt-out; milestones fire once, and the system seeds silently on first run so nobody sees a retroactive announcement. Controlled by tasksV2MilestoneNudgesEnabled (default ON) — the Milestone nudges row in the Settings Tasks section turns them back on. Every action button lands somewhere real: Create a project opens the rail's new-project button, View shortcuts the shortcuts overlay, Plan my day the Plan my day panel; when one cannot open anything you get a short "Couldn't open that" instead of silence.

Milestone nudges are desktop-app only. They arrive as in-app toasts over a channel the mobile and web bridge does not serve, so a phone or a browser on Omniscio web access never shows one and never sees the Milestone nudges setting do anything. Each nudge that is shown, acted on or switched off is counted in Tasks usage, so the whole shown-to-finished funnel is visible.

Your first time in Tasks

  • Welcome tasks. A brand-new account starts with four welcome tasks in the Inbox. The two that show off AI carry a short note, Tasks AI is not connected yet, with a Set up AI button that opens the page where you connect AI; the note disappears once Tasks AI can run. Rewording one makes it yours.
  • They clear themselves. Once you add 5 tasks of your own (command-line ones count), Tasks clears away every welcome task you have not reworded — including ones you gave a date, notes, files or a project to, and ones you checked off. They go to Recently deleted, where you can bring one back for 30 days, and the sub-tasks you added to one are kept. A note above the list carries a Clear these button with an Undo toast.
  • Empty Today. A plain line says how to add something without AI: type in the + box, or press your Add to Today key (Y unless you changed it) on any task; with no such key it names only the + box.
  • Your own AI key and model. The Daily check-in start screen says "Tasks AI uses your own AI key and model. You can add or change it any time." with an AI settings link.
  • The conversation follows the newest message in the Daily check-in: the thread scrolls to the bottom on open and on each new message.
  • Break it down and Plan my day open a chat. Each starts an agent session and switches the Tasks panel to its Sessions tab with that chat open (part 8, part 5).
  • Meet Tasks. Once per account (not once per computer), someone whose setup is finished, whose first day is over and who has not used Tasks yet gets a Meet Tasks inbox card. It is checked when the app starts and each time someone signs in, so a second account on the same computer meets Tasks too. With Tasks off the card has a one-click Turn on Tasks (desktop only; a "Tasks is on" toast offers Undo); with Tasks already on for the computer it says Tasks is already on and can be found in the sidebar. An account that already has tasks, or got the welcome tasks, never gets it.

Set up Tasks in one place

The four things Tasks depends on — AI for Plan my day, Caught, Memory, and the Tasks panel in Mission Control — live in the Set up Tasks card at the top of Tasks → Customize, each with a live switch, so you can see what is on and change it without hunting through Settings. The first time you open Tasks, Customize opens itself once with that card in view.

  • Turn on all switches on every row that is still off, in one click. Each row says in plain words what turning it on does; Caught and Memory say where your data goes and that they use AI.
  • The AI row tells you why AI is not running. It says whether AI is ready, still checking or unavailable. The Mission Control row says so when Mission Control itself is off.
  • It opens once per account. The first visit opens Customize itself; leaving it in any way means it never opens on its own again for that account. If all four are already on, nothing opens at all. It waits until Tasks is actually on screen, so it never pops up while you are elsewhere in the app.
  • Open it again any time from ⋯ More → Customize in Tasks, or Settings → Tasks → Set up Tasks → Open (that dialog renders the same four switches).

When something goes wrong

Tasks says what actually failed and keeps your work wherever it can.

  • Moving tasks into a brand-new project opens it only once the move worked: if nothing moved the empty project is removed and one toast says so, and if only some moved the project stays because it holds them. A failed master-context save reopens the editor with your text. A failed attachment upload names the file in one toast; a failed attachment load shows the error instead of the previous task's files.
  • "Share with agent" decides what an agent is handed, not what it can reach. Each attachment has one (off by default): a shared file's name, type, size and location go into the brief of an agent launched on that task (up to the 10 newest, plus a count of those left out). An unchecked file is not mentioned and not offered — but an agent session on this computer can still reach it, so keep a file off the task if it must stay out of reach. "Automatically share all files with agents" (off by default) ticks and locks every one.
  • A typo in the command line shows the parser's own explanation. Pasting an app link into a blank new row names the task after the link. A context note edited elsewhere writes nothing and offers Reload or Overwrite. A task you are typing in changed elsewhere shows "Another window or an agent changed '<task>'. Your version is not saved yet." with Keep my version and Discard my version. A task removed while you were typing says "'<task>' was removed, archived or moved elsewhere. Your unsaved wording was not saved." with Copy my text, so copy it before you move on.
  • Dismiss all in Caught dismisses every pending suggestion its prompt counted, rendered or not. A text-only edit the server refuses stays on screen with Save again; every other refused edit rolls back. A write that times out is treated as unknown, not failed: Tasks re-reads and says so, and a timed-out undo keeps the change, drops the step and reloads.
  • If Tasks cannot load at all, the full-page card says "Couldn't load your tasks" with Try again; the table below names what to check.
  • Never fail silently. Every failure says what happened and offers the next step: a failed refresh keeps the tasks on screen with "Couldn't refresh. Showing the tasks you had before." and Retry; a hand-clicked Get link title that finds nothing offers Try again while the automatic fill stays quiet; a fallen-back Re-rank, and the note under the Right now pick, names the real cause — Tasks AI off, a rejected key, used-up credits, or a plain outage — instead of one flat "AI is off"; a linked task that cannot open says "Couldn't open that task right now." instead of claiming it no longer exists; bulk actions announce success and clear the selection only after every row saved; an unsaved app link stays in the row so Save again saves it; a quick-add whose save timed out is not put back, so Enter cannot double-add; a deleted Super Prompt refuses the launch; the import preview lists up to 10 unreadable rows per field; the Mission Control suggestion retries its save once; and an AI reply that fails partway through the Daily check-in or Break it down conversation carries the same fix as one that fails at the start — Turn AI back on when Tasks AI is switched off, or Set up AI when no key is set, the key was rejected or the credits ran out.
  • No AI set up. Every backup suggestion says "Tasks AI needs a key before it can help, so this suggestion comes from your own task list." with a Set up AI button.
  • The task list never reloads itself in a loop: a change landing mid-reload restarts the read at most once, then waits.
  • A display crash stays in its own slot. Each embedded session chat has its own error boundary ("This chat couldn't be displayed.") while the session keeps running, and the meeting reminder, Attach-to-task picker and pop-up windows (task context, shortcuts guide, Task details) each show a small bottom-corner notice with a Dismiss. The AI panels — Daily check-in, Plan my day, About you, Customize, the task interview — show a notice naming the panel with Try again and Close, promising nothing about unsent text, and the sidebar's lists, the mobile project sheet and the Sessions start screen each have their own notice. Every boundary resets when the view changes, and the outliner's resets per project without remounting a healthy outliner.

Troubleshooting failures

Tasks' background work — AI calls, reminders, follow-ups, meeting alarms, catch scans, task agents, delegations, the daily check-in, the evening wrap-up and the tasks file your agents read — runs with no screen to show an error, and the task list itself can fail to open. Each row says what you see, the inbox card it raises, and what to check.

What you see Inbox card (dedup key) What to check, and what to do
"Couldn't load your tasks" fills the page none Check that Tasks is on (the list read is refused while it is off) and the database unlocked if encryption is on. Press Try again — a busy database usually answers on the next read — else restart and read the [IPC:tasks-v2:list] failure. A failed refresh with tasks on screen is the "Couldn't refresh" notice above
One AI feature keeps using its simpler built-in version Tasks AI keeps failing for <feature> (tasks-v2-ai-source-failing:<source>) after 3 of that feature's last 10 calls for one model fail Change that feature's model — other features succeeding points at its model or prompt, and a refused key, exhausted credits or a concurrency refusal never counts. It clears on that feature's next success
"Some task changes needed a second save" tasks-v2-write-heal Each heal saved the change, so nothing was lost; it clears on the first quiet check
"Task changes could not be saved" (urgent) tasks-v2-write-heal-failed A change you made in the last few minutes did not save even on the retry, so it may be missing. The card names the tasks affected: open Tasks, check those edits and redo any that are missing. If it keeps happening, report it from Feedback
"An unencrypted copy of your tasks is still on this computer" tasks-v2-plaintext-mirror-left Turning on Database Encryption or erasing your data could not delete every readable copy of your tasks (the tasks-v2.md mirror or an Export JSON folder), usually because another program has the file open. Close it (a sync tool, an editor, a File Explorer window) and restart — the app tries again at the next start. If the card comes back, delete the tasks-v2 copies by hand
"Task changes could not be saved" (critical, shows on day one) tasks-v2-write-heal-failed, with each affected task named in the card's fields The second save failed too, so a change made in the last few minutes may be missing from the listed tasks — press Open Tasks, check the edits made in that window, redo any that are missing, and report the metric= token from the [telemetry-health] log. This card does not clear itself: it stays until you dismiss it
A due, snoozed or follow-up reminder never appears none At most 5 go out per one-minute tick, so a backlog trickles — wait a tick. "Due-soon alert (hours before deadline)" at 0 turns due-soon reminders off; if it never arrives, restart the app
A meeting reminder never appears, or the meeting passed tasks-v2-reminders-stalled:tasks-v2-event-clock after three failed ticks An alarm only rings while the app runs, and one over 30 minutes past due is skipped rather than rung late; fix the [tasks-v2-event-clock] cause
"A task reminder keeps failing" A task reminder keeps failing (tasks-v2-task-reminder-failing:<taskId>), one card per task Fix the logged cause; it clears when that task's reminder goes through or the task is finished, deleted or snoozed past scanning
Session-end catches stop showing up in Caught Some finished sessions were not checked for follow-ups (tasks-v2-catch-scans-lost) after 3 lost scans Turn on "Catch follow-ups when a session ends" and include the session's project in any filter. A refused key is never retried. queue-overflow scans and scans left by an outage of the AI endpoint (a timeout or an open circuit breaker) return via an hourly sweep using "Look back for missed sessions" (3 days, counted from the account's own first Caught sweep: a computer remembers only the most recently swept account's starting point, so an account that signs back in after another one swept starts again from that point). A retries-exhausted scan comes back too: restarting the app clears those losses and the sweep queues the sessions again inside its look-back window, and a later check that goes through replays the ones the same run gave up on. Only stranded-backlog scans are dropped for good
"Task agents keep failing to start" tasks-v2-agent-launch-failing Three real failures in a row raise it (a launch limit or timeout never counts); it clears at the first agent that starts
"A task agent didn't start" / "Inbox Triage didn't start" / "Sorting your caught items didn't start" tasks-v2-launch-late-failed:<taskId> / tasks-v2-triage-launch-failed:<mode> The launch failed after the one-minute wait answered "still starting", and the slot is freed — start it again, and the card clears when it starts
A batch, whole-list, Plan my day or macro launch of agents did not start An agent didn't start (`tasks-v2-launch-kind-late-failed:<batch list
"Task reminders are not being delivered" tasks-v2-reminders-stalled:<serviceId>, one card per clock: tasks-v2-reminder, tasks-v2-followup-clock, tasks-v2-event-clock A tick where every reminder failed counts as one; three in a row raise the card, so fix the per-task failed lines above
The card reached your Inbox but your phone, email or text did not none Turn on the channel — "Email me task reminders", "Text me task reminders", "Send task reminders to my phone" — needing a signed-in agent address, Twilio, and a paired phone or Web Access subscription respectively; quiet hours and Do Not Disturb skip the phone silently, and a per-reminder-type channel choice wins. A card still held five minutes past the end of its hold, or pushed out by a 200-card backlog, never sends them.
No daily check-in, or a plain local summary instead of a conversation none Turn on "Daily check-in" (off by default) and set "Daily check-in hour" (8 by default) — it opens at or after that hour on a 15-minute check, and stops after three failed opens until tomorrow. A one-off AI failure still gives the local summary
No evening wrap-up note arrived none "Evening wrap-up note" (on by default) and "Evening wrap-up hour" (6 PM) check every 15 minutes, once a day, quiet on a day already closed out; the next check retries
Your evening close-out asks again as if you never closed out none The database, not the AI: the close-out is marked done only after your day's plan and results reach the day log. The [tasks-v2:distill] learning pass never blocks it
Your earlier tasks are missing, with "Some of your earlier tasks are hidden for now" tasks-v2-account-adoption-failed, only while signed in It gives up after five fast attempts and one slow re-attempt about ten minutes later. Restart the app — nothing is lost, and the card clears once the move succeeds
The tasks-v2.md file your AI agents read is stale The tasks file your AI agents read is out of date (tasks-v2-mirror-stale) The app is fine — GET /tasks-v2/markdown renders from the database — so trust the list over the file. A sync tool, antivirus or full disk is the usual cause: clear it, then change any task or restart. The write retries within seconds, and if the quick retries run out it raises that card and keeps trying every few minutes for a while before leaving the file alone until the next app start
A delegation fails or shows "timed out" Draft didn't finish: <task> (tasks-v2-delegation-failed:<taskId>), also pushed to your phone A session that gives up on a failing provider or a refused sign-in fails its delegation in about 10 seconds and is stopped; one stopped by hand, or one that crashed and may restart, is left to the 5-minute sweep, and no progress for 2 hours fails it with "The delegation session timed out without producing a draft. You can try again." A deleted or filed-away task fails with "Stopped because its task was deleted or filed away. Restore the task to delegate it again." Delegate again, or restore the task first

How Tasks AI spending is recorded

Every paid AI call Tasks makes is written to the app's spend record with a label saying what caused it, so Stats → Spend shows where your Tasks AI money went: catches from a finished session against that session as tasks-v2:catch, button-started calls with the feature's own label (tasks-v2:right-now, checkin-open, checkin-reply, plan-my-day, interview, plan-chat, arrange — which also keeps check-ins and Plan my day separate in the record), and the two background passes as tasks-v2:right-now-rerank and tasks-v2:distill. A call that timed out after it was sent is still recorded, because the provider may have charged for it; Do it for me is not in this list, because it runs as its own agent session. The labels only describe the spending. No Tasks AI call is ever refused for what was already spent: there is no Tasks daily budget.

Your tasks belong to your account

Tasks is a personal list. Nobody else can see or edit your tasks, and there is no way to share a project with another person; for shared team work, use Mission Control.

Tasks keeps each Omniscio account's tasks apart, along with its projects, check-ins and suggestions: signing out shows only the tasks you made while signed out, and your account's tasks come back when you sign in again. Most Tasks settings follow the account too, filed under whoever wrote them so a new account starts from the stock values: the follow-up catching switch and its guidance, the ended-session scan's project filter and look-back, every Caught source and whose items it brings in, Memory, the AI next-task pick, the daily check-in switch and hour, the evening wrap-up switches and hour, the due-soon alert hours, automatically sharing files with agents, your custom Today and Caught names, colors and icons, your shortcut overrides, your collapsed and pinned projects, and whether the welcome tasks were given out. The rest apply to everyone on this computer: Tasks itself, the reminder alerts switch and the email, text and phone reminder channels, Plan my day and its day-load labels, AI arrange and its model, Inbox triage, Break it down, and the milestone and looks-big nudges, the display options in Customize, the calendar switches, keeping Tasks loaded, trickles while Tasks is hidden, the right-now card, and the 30-day purge of deleted tasks. Each Tasks setting's own help entry says which it is.

Signing in moves tasks you made while signed out into your account, but only on a computer one account uses; on a shared computer nothing moves and each account sees only its own tasks. An account with no project at all gets its own Inbox on sign-in, and one whose projects were all deleted is not given a new one. Signing out throws nothing away: a task you were still typing is saved before the account changes, and a list still loading is discarded rather than drawn. Do it for me drafts are the exception: a draft still being written is stopped and marked failed, so delegate the task again after you sign back in.

Ask any agent to add a task

While Tasks is on, the agent sessions you start on this computer are told that "add this to my tasks" or "put this on my list" means your Omniscio Tasks list, and to file it there rather than in the agent's own private to-do list, which you never see. Whether it actually files there is up to the agent — that part is an instruction the app cannot force. What the app does enforce is confirmation and size: the first time an agent adds something it is asked to confirm, and only the exact same request again from the same session, within 10 minutes, adds it. A task line over 200 characters, or a Context note over 50,000 characters, is refused with a validation error and nothing is written — so if a capture is refused for length, ask the agent to shorten it. Sessions running on another machine over SSH, and Ask Omniscio, do not do this.

Handing a task to an agent and getting the result back

  • Start it from the Launch agent window. Highlight a task and press g or its Launch agent button, then pick Do it for me (draft) under Approach in the Start session window (it works from the task as written: anything you type in Message or change in the task message is not sent) — a single task only, and only while Tasks delegation (AI does the task) is on in Settings, Lab. The row shows Drafting… and then Draft ready — review; that chip opens Review draft with Copy, Discard (asks first; you can delegate again later) and Accept, and Delegation history lists earlier runs. A delegate works only inside its own scratch folder: a read anywhere else, or any web or add-on tool, asks you first even when you have already allowed that tool — so text the delegate was asked to work on can never read a private file or send anything on its own.
  • "Do it for me" is on by default for new installs; an existing install keeps the value it already had, and one that never saved the setting stays off.
  • Each handed-off task is researched, drafted or summarized. The AI may pick one per task — never over a choice you made yourself — shown on a chip you can change; you can also type delegate research, delegate draft or delegate summarize (or assign) on the command line. With none chosen, Do it for me drafts, and the saved note is labeled to match.
  • The draft agent gets the whole task: project background, open subtasks, notes, links, reference docs, inherited parent context, and any files you shared with agents.
  • You hear when a draft is ready: one inbox card, "Draft ready: <task>", with an Open in Tasks button, cleared by accepting or rejecting, and an accepted draft stays on the task as an "AI draft" note. The card follows the same rule as the row's Draft ready — review chip: it is cleared while Tasks delegation is off, and it comes back on its own when delegation is switched on again and that draft is still waiting.
  • Finished agents report back. When an agent chat started from a task ends or errors, its last message is saved as one "Agent result — <chat name>" note (capped, emails and phone numbers removed), updated if the chat resumes, with no AI call made; a chat started from a task also shows a task chip in its header that opens it.
  • Each project can remember its code project through the project options menu's "Run agents in" picker, used by default for single, batch and whole-project launches, with a deleted code project falling back to the Tasks workspace.

Try Mission Control

While Mission Control is off, the Tasks header carries a Try Mission Control button on desktop, and after real planning use a one-time Ready for bigger projects? card appears under the header; both disappear once Mission Control is on. The rules for the card and the move are on Moving from Tasks to Mission Control.

Context attachments

A task's Context pane attaches the material it needs — a Session (a past chat), Project, another Task, a Project item (a Mission Control board item), a File, a Link, a Note or a ContextDock item — and a context item can cascade to sub-items. Clicking one opens it; the Project item option needs Mission Control on. The other direction lives on the item: its Links tab lists linked tasks under a Tasks heading, and a finished task drops off until you restore it. See Context attachments.

Editing a context item's label or note is non-optimistic: nothing is shown as saved until the server answers, and the TASKS_V2_CONTEXT_CHANGED push reconciles the open pane. A failed save must never be a dead button. The store therefore raises no toast of its own — it hands the dialog the humanized reason so the dialog shows one message in place ("Couldn't save that context."), and a save whose outcome is unknown (a transport timeout) says it may still be completing rather than claiming it failed.

Pop out into its own window

Tasks can open in its own desktop window, so you can keep your tasks on-screen beside other work. Two ways in: ⋯ More → View → Pop out in the Tasks header (desktop only), or right-click the "Tasks" row in the sidebar → Open in new window.

The popped-out window is the whole Tasks panel — same list, projects, keyboard, AI surfaces and per-task agent chat — and stays fully live, with both windows showing each other's changes. It remembers its size and position, follows your light/dark theme, wears Omniscio's own themed title bar matching the KMS pop-out rather than a plain native frame, and opening it again just focuses the window already open. The main window keeps its Tasks view too, and the Pop out row is hidden once you are inside the pop-out.

Keyboard: two modes (writing vs command)

Tasks is built to be driven entirely from the keyboard, borrowing vim's two modes. There is no mode badge — rendered markdown on the focused row means command mode, a raw-text editor with a caret means writing mode. The focused row carries a strong cursor highlight (a solid accent left bar, a faint tint and hugging ring, distinct from a multi-selected row's fill and glow) and turns green while you type. Escape releases the row, and a following Shift+Tab leaves the list; arrowing back in lands on the row you released.

A small mode label at the row's end says what your next key will do: "Shortcuts on · Space to type" (Space, or F2, starts typing in the row; Enter adds a new task below) or "Typing · Esc for shortcuts". It appears for a few seconds after you land on the row or change mode, then steps aside so the row stays quiet. Turn it off with Show mode label in Tasks → Customize or Settings (on by default, desktop only).

Opening Tasks puts the keyboard cursor on the first task (desktop only, including a kept-warm panel and the pop-out), so the arrows / w-s / j-k work immediately, and it never steals a focus you placed inside the list. The highlighted row shows its action controls, the same ones hover reveals.

Clicking or tapping anywhere on a row highlights it, and a double-click edits. Finishing the highlighted task moves the highlight to the next task inside its project — below, else above, else the first task of the next project — and snoozing moves it the same way; a bulk snooze leaves it put. The list scrolls to keep the focused row in view, and a row you are moving stays on screen on every reorder key — the one-hand keys (Shift+W/Shift+S, Shift+A/Shift+D) and the arrow route (Alt+Shift+↑/↓, Tab/Shift+Tab, Alt+Shift+←/→) alike. The indent step is 28px on desktop and 14px on a phone from a fixed 8px gutter; on a phone the visible indent stops growing past six levels so a deep subtask keeps room for its buttons, while the real depth survives for nesting and moves.

  • Writing mode — you land here whenever you create a task or open one for editing. Enter commits and returns to command mode (so does Esc) and never leaves a stray empty task; Cmd/Ctrl+Enter commits and opens a fresh row below in edit mode for rapid entry. (Finish the task itself — done and archived — is x / e / the checkbox / Delete.)
  • Command mode — the row shows rendered markdown and single keys act on it. Space edits it, Enter adds a new task right below, Shift+Enter one directly above. A new or empty row opens straight in writing mode.

While you write, a line break added with Shift+Enter or pasted in stays real, and a formatted paste becomes plain lines on landing. A task holds up to 10,000 characters, keeping your existing text and dropping only the overflow from what you just typed. Due and snooze chips keep themselves current while Tasks is open.

Four panel-wide behaviours round it out: switching views paints the last known load and reconciles behind it, so a genuinely loading tree shows a skeleton rather than "No tasks yet", and Today's quick-add clears its box on the click with duplicate pastes deduped; Tasks reopens where you left it, unless someone else signs in; the notices above clear themselves, a reminder failing three times for one task raises one notice keyed per task, and the launch limit is 30 task agents a minute and 200 a day (see Per-task agent chat); and the list never reloads in a loop, while an approved agent launch that never started can be approved again.

Speed cost

Measured November 6, 2026 on a test copy of the app (the cloud perf runner, a Linux machine), with Tasks off and then on. In the on run Tasks was switched on in Settings → Features, 4 projects with 200 tasks were added, Tasks was opened once, closed, and opened again, and then the app sat on the File Converter screen for 30 seconds while it was measured.

Measure Off On Cost of having it on
Background activity 14.7 CPU-s/min, 52 calls/min 20.8 CPU-s/min, 26 calls/min +6.1 CPU-s/min (about a tenth of one processor core); fewer calls to the back end
Typing delay median 14.7 ms, p95 32.6 ms median 10.6 ms, p95 28.5 ms none measurable
Freezes worst stall 65 ms, 0 long tasks worst stall 45 ms, 0 long tasks none
Memory 1,319 MB 1,984 MB +665 MB (about half again)
Startup 11.1 s 11.8 s +0.7 s

Tasks-only measures: time to open Tasks is 0.46 s the first time after the app starts and 0.17 s when you come back. Download size is the same with Tasks on or off, because Tasks is part of every copy of the app; the Tasks screen's own code, loaded the first time you open it, is 23 files and 816 KB (258 KB compressed).

Verdict: typing, freezes and startup barely change, but memory (+665 MB) and background activity (+6 CPU-seconds a minute) are worth looking into before Tasks reaches more people. Not measured: Tasks' AI helpers and the agents you hand tasks to (no AI calls ran), the phone, an account much larger than 200 tasks, Windows and macOS, and how much of the always-downloaded main app code belongs to Tasks. After the first open Tasks stays ready in the background (on by default), so the on numbers include that.

This page continues across nine part pages, each carrying one stretch of the detail: part 2 is the rebindable keyboard-shortcut table; part 3 covers Caught and its AI suggestions, emailing tasks in, inbox triage, cut and paste, duplicate, links and archive with universal undo; part 4 covers reminders by email, text and phone, the waiting-on follow-up clock, outline numbers and the Liquid Glass skin; part 5 covers rich fields and fast capture, emphasis, natural-language and voice capture, sort and filter, Right now, the daily check-in and Plan my day; part 6 covers the Today plate and projects with master context; part 7 covers find, bulk select, repeating tasks, the giant all-projects view, pinned columns, trickle and export; part 8 covers Break it down, per-task agent chat and the Sessions tab; part 9 covers context and file attachments; part 10 covers the markdown mirror and control-server routes; and part 12 covers closing out the day, About you, delegation, follow-ups, meetings and importing.

Beside it, Tasks helpers covers the urgent and important groups, "Looks due" chips, meeting items in Caught and the close-out review, and Moving from Tasks to Mission Control covers the Try Mission Control button and the move. The Tasks overview is the one-page tour on the help site. The earlier personal outliner has its own page, kept only as a redirect, and agents started from a task are ordinary sessions, so the page on starting a session applies to them too.

Last verified 2026-10-06