Tasks (markdown outliner, in development) (part 5)
Every task can carry a little more than its text — all optional, all shown as small chips on the row only when set, so rows stay clean. An Importance dial runs None / Low / Medium / High. An optional Urgency dial appears beside it when Urgency is turned on in Tasks Customize or Settings → Tasks · Caught → Tasks display; it is off by default. A task can also have its own start date — "when I plan to start this," separate from the due deadline. The page goes on to cover fast capture, sorting and filtering, the "Right now" view and its daily check-in, and Plan my day.
On this page
What it is
This is part 5 of the Tasks (markdown outliner, in development) page. It carries the next stretch of the material on that page, moved here because a single page is capped at 40,000 characters.
Where to find it
Reach this part through Tasks (markdown outliner, in development) — it lists every part and explains where the feature lives in the product. Everything below is reached from the same place.
How it behaves
Everything below is the behaviour, detail and edge cases that belong to this stretch of the Tasks (markdown outliner, in development) page.
Rich fields & fast capture
Every task can carry a little more than its text — all optional, all surfaced as small chips on the row only when set (so rows stay clean):
- An Importance dial — None / Low / Medium / High. Stored as a real column (
importance, integer 0–3) so a later "what should I do right now" view can sort on it quickly. An optional Urgency dial appears beside it, and as a chip on the row, when Urgency is turned on in Tasks Customize or Settings → Tasks · Caught → Tasks display; it is off by default. - A scheduled start date (
start_at) — "when I plan to start this," separate from the due deadline. A start can never be later than the deadline. In the details pane (and theddeadline picker), setting a deadline earlier than the start moves the start back to the deadline, and setting a start later than the deadline moves the deadline out to that start, so check the deadline after pushing a start date back. - In the details pane, the Due, Start and Follow up date boxes come from your computer, so they can show its language and date order. The date and time you picked are repeated just above each box in Omniscio's own language (for example
23.10.2026, 09:00in German), so you can always check the value. - An estimated time (
estimated_minutes) — how long the task is expected to take. Type it naturally in the details pane —30m,1h,1h 30m,2h, or a bare number for minutes (90) — and it shows back tidy (1h 30m) both in the field and as a ⏱ chip on the row; empty clears it. Text it cannot read ("banana") shows "Couldn't read a time from …" under the field when you leave it, and the details drawer will not close over it: X, Done, Escape and Ctrl/Cmd+Enter keep the drawer open, move you to the field, and change the message to "…Type a time like 30m or 1h, or clear the field to close." (screen readers announce it). That matches how a half-typed date already blocks the close. Bounded to 1 min…100 h. A recurring task carries its estimate to the next occurrence (a stable property, unlike the start date). Parse/format live in one shared helper (src/shared/tasks-v2-duration.ts). The editable field (the details pane and thetpicker) always prefills and settles to the plain1h 30mtoken — the same form the parser reads — even when the interface language is not English, so an untouched field never round-trips into a different estimate on blur; the localized label is used only for read-only display (the row's ⏱ chip, the presets, the AI panel). A time it can read settles to what was actually stored (1.5hshows back as1h 30m), and a blur that would store the same number writes nothing at all. - Tags — free-form labels, stored in the task's
metadata.tagsJSON (no column). Saving them leaves every other key already on the task's metadata untouched; onlytagschanges. - An Energy level (
energy_level) — High (needs your sharpest focus), Medium (steady attention), Low (fine when you're tired), or Not set. Shown on the row as a ⚡ "High energy" / "Medium energy" / "Low energy" chip. - A Context tag (
context_tag) — a short word for where or with what you can do the task, such asphone,errandsorcomputer(up to 200 characters; a leading@is dropped, and an empty box clears it). Shown on the row as a 📍 chip. This is a single label, separate from the task's context items (notes, links, files) in the Context pane.
A row shows at most six chips; any more fold into a +N badge (hover it for the count). Open Task details to see every field.
Energy and Context sit in the details pane's Effort & priority group, under the estimate. Both are restored by Ctrl/Cmd+Z and given to AI Arrange on Today (each task line in its prompt gains energy=… and context=… when set) so it can match the work to your energy.
You edit all of these in the details pane — opened from the row's sliders button, Cmd/Ctrl+;, or by clicking a due / importance / tags chip on the row — or set importance fast with the p key (a quick None / Low / Medium / High picker). The pane's sticky header shows the task's name above the Details and Context tabs, so you always know whose details you are looking at, even after jumping from one linked task to another: a long name is cut short with an ellipsis and shown in full when you hover it, it changes the moment the task is renamed, a blank task reads "(untitled task)", and screen readers announce the pane by that name. Each field saves automatically as you leave it. Snooze in the details pane opens the same universal snooze palette as the h key (plus a Clear), not a raw date box. Every typed field saves the same way — write only when the value actually changed, and a re-typed identical date writes nothing — so no field invents its own save ritual, and the repeat count the picker offers and the number written on blur can never disagree.
Emphasis — bold, highlight & star a task
Three lightweight ways to make an individual task stand out, all per-task and all separate from the Importance dial above (which is for prioritization — emphasis is purely visual):
- Star — flags the row with a ⭐ marker (a quick "favorite", distinct from priority).
- Bold — renders the whole row's text in bold.
- Highlight — gives the row a highlighter background in one of five colors (yellow, green, blue, pink, red). The colors are theme-aware — translucent, so they stay legible in both light and dark mode.
Set them three ways: in the task's details pane (a "Style" row beside the priority dial), as quick toggles in a row's right-click / ⋯ menu (Star, Bold, and a color-swatch picker), or by typing into the : command line — star / unstar, bold / unbold, or highlight yellow (or green, blue, pink, red). A bare highlight, none or off clears the highlight, and a color it does not offer (highlight orange) is refused with a message naming the five instead of quietly clearing what the task had. Capitals and a closing period are fine: Highlight Yellow. still turns it yellow. Each saves instantly, shows in both the giant "All Projects" view and a single-project view, and Ctrl/Cmd+Z undoes it like any other edit. (There is no single-key shortcut for emphasis yet — that's an additive later step via the rebindable command registry.)
These are presentation only: they change how a task looks and drive no sorting or prioritizer, and — unlike Context and tags — they do not appear in the tasks-v2.md mirror, so your export stays byte-identical. The controls live in one shared details pane and row menu, so the two views cannot diverge.
Natural-language capture
The dedicated "Add a task" box was removed from Tasks — you add tasks via each project's "+" (a master-view group header), the per-project outliner's always-present "Add a task" row, the empty-state's focused first-task field (on an empty list the cursor lands in an "Add your first task…" box — just type and press Enter; see Empty list opens ready to type), or Enter on an existing row. The "Add a task" row and the other new-task boxes turn natural language into a task — type something like:
call vendor due friday #work !high
…and it parses the due date ("friday" → the upcoming Friday; dates lean future), the tag (#work), and the importance (!high) — also start <date> and an estimated time (~30m, ~1h30m, ~2h). This same parser runs when you commit a FIRST task, so the empty-state's first-task field turns call vendor due friday #work !high into a structured task rather than storing the sentence as the title — a cold user gets the rich capture without first learning the : command. The ~ sigil sets an estimate and nothing else (~30m, ~1h30m, ~2h); no typed mark sets urgency, which comes only from the spoken words "urgent" and "not urgent" — and these boxes do save it. A date is never silently guessed (a wrong due date is a missed deadline). A date word that names the task stays in the name: in the new-task box a date word right after a bare "the" ("Plan the weekend trip", "Book the friday meeting room") is part of the title and sets no date, while a due keyword in front of "the" takes the whole phrase ("Call mom on the weekend" saves as "Call mom", due Saturday). "for" is a due keyword too, so it leaves with its date instead of dangling: "Pack the gym bag for tomorrow" saves as "Pack the gym bag", due tomorrow (BUG-B031). The natural-language dates come from the chrono library, loaded on demand so it never weighs down startup. The pure parser is src/renderer/src/features/tasks-v2/parse-quick-capture.ts (shared with Tasks). Keywords and repeats are English. The due/start keywords ("due", "by", "on", "for", "start", "from") and the recurrence phrases ("daily", "weekly", "every monday") are read in English only; the dates themselves follow your app language, as the next paragraph explains.
Date words follow your app language. When the app is in German, French, Spanish, Portuguese, Dutch, Italian, Japanese, Chinese, Russian, Ukrainian or Swedish, the capture also reads that language's date words ("morgen", "vendredi", "mañana", "明日", "завтра"), with English still understood as a second pass, so English dates keep working in every language. The small keywords in front of a date (due, by, on, start, from) stay English, and the language is picked when the date reader loads, so a language change applies from the next load. The Tags field in a task's details accepts the ASCII comma plus the Chinese, Arabic and Persian commas and semicolons as separators. The box holds up to 500 characters, a tag typed twice is kept once (capitals ignored), and you can type , to keep a comma inside one tag.
The : command line reads the same kind of words but applies them to the task you have highlighted — it never turns your words into a new task. Typing the line above there would give the highlighted task a Friday due date, the #work tag and High importance, and the words "call vendor" would be ignored.
New-task boxes read what you type (labels you can remove)
The three places you type a new task — the "Add a task" row, the empty list's "Add your first task…" field, and the Today view's add box — read the same things as you type and show each one as a small label under the box:
Call dentist tomorrow at 3pm #health→ labels Due Tomorrow, 3 PM · #health, saved as the task "Call dentist"
- What it picks up: a due date (
friday,tomorrow at 3pm,due oct 20), a start date (from monday), priority (!high,!medium,!low), a repeat (every monday,daily,monthly,every other friday,every monday and thursday,the last friday of the month,quarterly, with an optionalafter I finishanduntil dec 31— the full list is in Repeating tasks), an estimate (~30m,~1h30m), and#tags. - Tags stay within what can be saved. A
#taglonger than 200 characters, or any tag past the 200th, stays in the task name instead of becoming a tag, so one pasted#tokennever makes the whole save fail. - Enter saves exactly the labels you see. Remove a label with its ✕ (or middle-click it) before pressing Enter and that field is not saved — its words stay in the task name instead (remove the date above and the task is "Call dentist tomorrow at 3pm").
- It is careful with dates. A date is read only when it names a day, a weekday or a clock time, so "Fix bug in May release" and "Plan the week" stay plain names. A number after
#is never a tag (Buy #2 pencilsstays whole), and a~that carries no time (~high) stays in the name, because~only ever sets an estimate. Priority also takes a number from!1to!3; any other number (!0,!4, or a decimal such as!2.5) is not a priority and stays whole in the name. - A day with no time follows one rule everywhere — these boxes, the
:command line, theddeadline picker, plan chat and an imported CSV all agree down to the same instant. Today's date (today,oct 16typed on Oct 16) is due at the end of today, never "right now" and never next year. Any other day (tomorrow,friday,in 3 days) is due at the end of that day (the last moment before midnight), so it is not overdue during its own day. A part of the day keeps its hour (tomorrow evening,friday afternoon), and a part already past today (tonighttyped at 11:30 PM) becomes the end of today. One deliberate exception is a month and day that has already passed this year (oct 20typed on Oct 27): theddeadline picker keeps it in this year, because a deadline is allowed to be overdue, while the new-task boxes and the:command line read the next time that day comes round. A typed year is always honoured, in the past or the future. - Day counts and months use the calendar.
in 1 weekis the same clock time seven calendar days later, even across a daylight-saving change, andnext monthon Jan 31 lands on Feb 28 (or 29), not in March. - A time that daylight saving skips (
tomorrow 2:30amon the spring-forward night) is saved at the shifted time (3:30 AM) and the label shows the daylight-saving warning, instead of leaving the words in the name. - A repeat with no date gets the usual first due date (the end of tomorrow), the same as the command line. When the repeat names a day (
every monday), the first due is the first matching day on or after tomorrow, still at the end of that day. - Today box meetings: a clock time ("Call Sam 2pm") still makes a meeting on your agenda, now shown first as a Meeting · Today, 2 PM label; remove that label and it becomes a plain Today task. A date without a time ("Pay rent friday") stays a Today task with that due date.
- Labels wrap under the box on a phone, and each ✕ is a real button you can reach with Tab. Dates appear a moment after you start typing (the date library loads on first use); if you press Enter before a date label shows, the date words simply stay in the name.
Talk your tasks in (voice task entry)
You can dictate into the same three boxes with whatever voice typing you already use — Omniscio's own dictation, Windows voice typing (Win+H), Mac dictation, or your phone keyboard's mic. Tasks has no mic button of its own and starts no recording, so there is nothing extra to turn on and nothing extra to pay for.
"Call mom tomorrow, high priority. Buy milk, tag groceries. Next task book flights, takes thirty minutes." → three tasks: Call mom (Due Tomorrow · High priority) · Buy milk (#groceries) · Book flights (Estimate 30m)
- Words you can say instead of symbols (only while Omniscio's language is English — in any other language these English words are left in the task name, and the typed
#tag/!high/~30msymbols still work): "high priority", "highest priority", "top priority", "priority high" or "important" (high), "medium priority" or "priority medium", "low priority", "lowest priority" or "priority low", "not important", "not high priority", "not highest priority" or "not top priority" (low), "urgent" (sets the urgency rating to High) and "not urgent" (urgency Low), which never change importance and are saved even while the urgency dial is hidden; once the Urgency dial is turned on, urgency also counts in the ranking and in the urgent and important groups, "tag health" or "hashtag health" (a #health tag), and "takes thirty minutes" / "takes an hour" / "takes 2 hours" (an estimate). Typing these words works the same way. Capitals, the commas a dictation tool adds and numbers spoken as words are all fine, and a closing period never ends up in the task name. The:command line reads the same priority words, except "urgent" and "not urgent", which it does not understand. - Several tasks at once: say "next task" (English only), or just finish a sentence and start another, and the box shows a numbered list of the tasks it will add, each with its own labels. Click Keep as one task to make it a single task again (Split into separate tasks undoes that). This works for typed sentences too. A period inside "Dr. Smith", "3 p.m.", "2.5 hours" or "U.S." never splits a task.
- Nothing saves until you press Enter (or tap the add button on a phone). Enter saves every task on the list, in the order you said them, with exactly the labels shown. If you press Enter while your dictation tool or phone keyboard is still putting words in (or a split second after a phrase ends), Tasks waits for it to finish (for Omniscio’s own dictation, until it has pasted the words) and then saves the whole sentence once, so half a sentence never saves; Escape cancels that waiting save.
- Labels keep up as you talk: they appear and update as the words arrive, and a label you removed with its ✕ stays removed while you keep talking.
- Good to know: an ordinary phrase can look like a spoken word — "read the important email" shows a High priority label, and "review next task list" splits — so glance at the labels and the list before pressing Enter; removing the label or clicking Keep as one task puts it right. A phrase you removed is removed on every line that says it. A word you format is left alone:
**Important** meeting with _client_saves exactly as typed, with no priority, because a spoken word or a date touching*,_,~or a backtick counts as formatting. The same word without the markers still works.
Smart capitalization
As you type a task, the first word and the start of each new sentence are auto-capitalized — the same "smart" feel as the KMS notes editor. It is intelligent about what to leave alone: URLs, file paths, abbreviations (Dr., etc., i.e.), decimals (3.14), and U.S.-style initials all stay exactly as typed (as do already-capitalized words). It runs live as you finish each word in the per-project "Add a task" row (caret-safe — the transform only flips one letter's case in place, so your cursor never jumps to the end) and on blur when you edit an existing task's text (the contenteditable's caret is fragile, so the edit path tidies up only when you click away). A task you capture from the Ctrl+Space Quick Launch modal's Task tab is capitalized too — its saved text comes out exactly as if you'd typed it into the list. It is setting-gated — Settings → Lab → "Auto-capitalize task titles" (autoCapitalizeTasks, default on); turn it off and your text is stored exactly as typed. The rules (the URL/abbreviation/decimal/initials matrix) live in one shared module, src/shared/smart-capitalize-core.ts, used by Tasks and the Quick Launch task capture, so they can't drift. The Vault notes editor does not capitalize: what you type there is what the note saves. The plain-string wrapper is src/shared/smart-capitalize.ts; invariant smart-capitalize in the contract locks that the transform is length-preserving and never flips a row blank.
Whitespace trimming
When a task is saved — whether you create it or edit an existing one — any leading or trailing whitespace is stripped (so " buy milk " is stored as "buy milk"). Spaces between words are left exactly as typed; this is a trim, not a re-spacing. Unlike smart capitalization it is not setting-gated — it always applies, and it happens at the save (persistence) layer, so every entry point gets it: the outliner, the : command line, the Quick Launch capture, and the CLI routes. It never leaves a task blank — a title that is only whitespace is treated as an empty edit and dropped, exactly like the "no blank tasks" rule. The invariant auto-trim in the contract locks this for both the create and edit save paths.
Add tasks at the top or bottom
New tasks added via the always-present "Add a task" row land at the top of the list by
default — the newest task sits above the existing ones. You can flip this under Settings →
Tasks · Caught → Tasks display → "Add new tasks at" (tasksV2QuickAddPosition, an enum 'top' | 'bottom', default
'top'): choose Bottom to append new tasks to the end instead (the pre-setting behavior). Because the "Add a task" row sits below the list, a task added at the top of a list you have scrolled down would land out of sight. So in that case a short "Added to the top of the list" line appears under the row for a few seconds, with a Show button that scrolls to the new task; if a filter is hiding that task, Show clears the filter so the row actually appears. A task you add while a filter or a find search is on is still saved, and when the current filter hides it a "Added, but hidden by your current filter" line appears under the row with a Clear filters button, so the box never just swallows what you typed. Your typing stays in the row (TaskV2AddedToTopNote.tsx). The
control shows only when the Tasks Lab feature is revealed. Both "add a new task" affordances
honor the setting — the "Add a task" row AND the "+" on a project's header in the all-projects
view — so a new task lands in the same spot however you add it (before this, the project-header "+"
always appended, so a top-preferring task landed at the bottom there). The header "+", the project
menu's Add task and a new project's first task all put your cursor in that project's "Add a
task" row, so dates, tags, estimates and repeats you type there show as labels and are saved
(before this, they opened a blank row that saved the whole typed line as the name). The POSITIONAL edits keep
their own explicit positions and are unaffected: inline Enter (new sibling below), Shift+Enter
(insert above), and multi-row paste. Under the hood top reuses the same insert-above
mechanism as Shift+Enter: it resolves to a beforeId of the first row of the target list
(scoped by list, since the giant view interleaves every project's rows in one order), so no backend
or schema change was needed.
Quick capture from anywhere (Ctrl+Space)
The global Quick Launch popup's Task tab captures a title straight into Tasks — filed into the project you pick in the tab, or into Inbox when no project was picked or the one you picked no longer exists. The Task tab works only while Tasks is on: with Tasks off it is hidden, and there is no fallback to the old v1 task list, which has been removed. It's the "catch a thought from anywhere, triage later" entry point, saved in the main process via the QUICK_LAUNCH_CREATE_TASK router. It reads what you type exactly like the new-task boxes above — due and start dates, !priority, ~estimate, #tags and repeats, shown as the same labels under the box (each removable with its ✕) — and it splits several sentences or "next task" into separate tasks (up to 20; with more than that, the whole text is saved as one task, exactly as typed). A Due date you pick under Add details beats a date read from your words (on every task the box splits into), and Send later beats both: the task is snoozed until that time and keeps no due date, so it is never saved with two conflicting dates. The Find Task tab searches your open tasks and opens one in Tasks. See Quick Launch modal.
Inbox triage AI. When unsorted Inbox tasks exceed a threshold (10), a triage card surfaces and spawns a paid AI session that proposes filing each unsorted task into your lists and projects. The feature needs Tasks itself to be visible (the Lab switch) AND tasksV2InboxTriageEnabled (default ON): while Tasks is hidden, the main process refuses every triage spawn, reports an unsorted count of zero, and the inbox card neither loads nor shows. It counts toward the daily Tasks launch limit (200 per day). The same switch also gates the Caught lane's Triage this for me sessions, which start through the same main-side handler as a different mode.
Other features can add tasks too. When you start a study session on a pack in Flashcards, it adds a "Review Flashcards Pack" task to your Inbox and finishes it for you once you have graded the pack's cards — see Flashcards.
Sort & filter
The header's Filter funnel opens a popover that lets you sort the view (Manual order, Importance, Due date, Start date, Newest first / Oldest first by when a task was created, or Recently changed / Least recently changed by when it was last edited) and filter it (minimum importance; by Due — overdue, today, this week, or no date for tasks with no due date; by tag — a tag matches however it was saved, ignoring capitals and a leading #, and the tag list shows each tag once; a tag chip shows as on whichever form the saved filter holds, and one click turns that tag off; by Energy — High / Medium / Low; or by Context, which lists the context tags your tasks already use and appears only once at least one task has one, and whose matching ignores capitals, accents and a leading @); a dot on the funnel marks an active sort or filter (any sort other than manual counts). The sort and filters, Energy and Context included, are remembered across reloads and windows. Both are a view overlay only — they never rewrite your hand-arranged order, and filtering keeps a matched task's parent rows so the tree stays readable. There is no urgency sort or filter: urgency is a dial in the details pane and a chip on the row, not a way to order the list. A task whose parent row is not in the list you are looking at — a parent that lives in another list — is shown at the top level rather than dropped, the same rescue the all-projects view applies per list.
Which views have it. The header funnel and Find appear in the all-projects view and in a single project, which share one sort and filter. Today has its own sort and filter bar above the plate, remembered separately, so a filter you set on a project never hides tasks on Today. Caught and Archived have neither. Each view also follows a change you make in another Tasks window, such as the pop-out.
The dot on the funnel also lights for an active sort: any sort other than manual counts.
Saved views ("smart lists"). Under the sort and filter controls, the same popover has a Saved views section for bringing back a combination you use often (say, "High priority"). Set the sort and filters you want, type a name, and press Save view (or Enter): it appears as a chip. Click a chip to apply that view — it replaces the sort and filters you had — and click the highlighted chip again to go back to manual order with no filters; the × on a chip deletes that view. Saving under a name you already used, in any letter case, replaces that view. Once at least one view exists, typing a name while the sort and filters are at their defaults keeps Save view disabled and shows Pick a sort or filters above first. under it, so the disabled button always says why. A view remembers the sort, the importance level, the tags, the due-date filter and the Energy and Context filters, never the find text, and you can keep up to 20. Views are kept on this device only, are cleared if the signed-in account changes on a shared computer, and are on-screen only: there is no command-line route to list, create, apply or delete them.
All of this is additive and stays behind the Tasks Labs gate. (Dedicated link / document fields live in Context attachments — URLs, notes/markdown, attached files, and linked ContextDock bundles all attach there.)
Prioritizing AI — "Right now" + the daily check-in
"Right now" is a card at the top of Today, and it never spends on its own. The header carries no "Right now" pill; a slim Right now card sits at the top of Today and reads only the stored pick (
fetchRightNowCached, thecacheOnlyread ofTASKS_V2_RIGHT_NOW): the AI's cached verdict when one is fresh, otherwise the free local pick. The paid ranking runs when you click Re-rank (labeled as using AI), or when you press the What should I do next? card on an empty or finished Today list, which asks the AI for a fresh pick when the saved one is out of date. Both run only while the AI next-task pick switch (tasksV2RightNowAiEnabled) is on; with it off, Re-rank refreshes the free local pick. Turn the card off with Right now card in Tasks Customize (tasksV2RightNowCardEnabled, default on). The daily check-in and break it down are unaffected; each runs on a single click (no confirm pop-up).
Tasks can tell you the single highest-value thing to do next and run a short daily check-in. It reasons over the rich fields above (importance dial, due/start dates, and each task's time estimate when set; urgency too, weighing any value the task carries once the Urgency dial is turned on) plus how much context a task carries. The estimate is context it may weigh (a quick win vs. a long haul) — in the check-in it surfaces a rough total only if you ask about time or the day's focus clearly runs long, never unprompted.
Meetings are not work to rank. A meeting's time is when it starts, not a deadline, so the scorer leaves meetings out: a past meeting never shows up as overdue work in Right now, the daily check-in or the Due-soon card, and a list with only meetings left reads as all caught up. "Due today" means today on your own calendar — a task due tomorrow at 8 AM reads as "this week" when you check in at 9 PM, everywhere, including what the AI is told.
It's built to be cheap. A free, instant local scorer is the engine — it ranks your tasks with no AI at all (src/shared/tasks-v2-priority.ts), and that ranking always shows immediately. A cheap AI layer then refines it. When two tasks score the same, the tie follows the order you arranged your projects in the sidebar, then the position within the project — never the project that happens to hold the lowest internal ordering number. The AI's verdict is cached and only recomputed when your tasks actually change (a new task, a changed dial/deadline, a completion), so it doesn't spend tokens on every screen refresh. If the AI is turned off or unavailable, everything quietly falls back to the free local ranking. A rejected key or used-up AI credits are named as such, with a Set up AI button, and when Tasks AI is turned off, Right now, the check-in, Plan my day and Break it down all say "AI is off, so this uses your own task list. You can turn AI back on in Tasks settings." with a Turn AI back on button that opens the AI switch, with no AI call. With no AI set up at all (no key and no funded credits), the same surfaces plus the "What should I do next?" card and the arranged-by-priorities notice say "AI is off, so this suggestion comes from your own task list." with a Set up AI button that opens Settings → Accounts, whose Who pays card lists every way of connecting AI (a Claude login, a Codex or Grok account, Omniscio credits, or a key of your own) rather than assuming the key box, so the backup never looks like a real AI answer; the Right now card keeps that label until any Tasks AI call succeeds.
The AI never sees your whole list. Right now hands it the top 30 tasks of the free local ranking, so a task further down can only become the pick once it rises into that group. The daily check-in (and the single-call Plan my day engine the command line starts) is shown your overdue tasks, the tasks due today, the top 5 tasks of the free local ranking and the tasks you are waiting on, at most 12 of each with a count of the rest, plus how many tasks you have open. Each task reaches the AI as its first 140 characters; a longer task is cut only in what the AI reads, never in your own copy.
The two welcome tasks that show off AI say why it is off, and open the page that fixes that reason. "Use Plan my day to let AI organize your tasks" and "Check the Caught queue for AI-suggested follow-ups" carry a small chip while Tasks AI cannot run (TaskV2StarterAiNeed.tsx, whose answer comes from tasksAiAvailable in tasks-ai-blocker.ts). The chip never assumes a key: while no AI is connected it says "Tasks AI is not connected yet" and its button opens Settings → Accounts, the same Who pays page every other Tasks AI prompt uses. It renders nothing at all once Tasks AI can run, and nothing before the availability probe has answered, so it can never flash in front of someone whose AI already works.
The Today band always leads to Today. The Today band at the top of the master view (and of the Mission Control Tasks side panel) always shows Open in Today, even while Today is empty, and its Today heading is a button that opens the Today view too. While Today is empty the "Plan my day in a session" button sits beside Open in Today.
A switched-off Plan my day says so, without waiting to be clicked. The Today band at the top of the master view shows the "Plan my day in a session" button while the switch is on; while tasksV2PlanMyDayEnabled is off it shows a plain "Plan my day is off." line with a Turn on button (MasterViewPlanMyDayAction.tsx → PlanMyDayOffNotice.tsx) that flips the setting and starts nothing. The confirm dialog still guards the other road — pressing Plan my day while it is off — because that person did not ask to turn it on.
- "Right now" (a card at the top of Today) — the one next action, a due chip (Overdue, Due today …), the AI's one-line reason (only while the card shows the task the AI picked, not after Not now moves on), and the next few tasks on the shortlist. Clicking the pick shows that task, Not now cycles to the next-best, Re-rank asks the AI for a fresh take. The card hides itself when nothing is actionable. It's assistive — it suggests, you decide.
- "Right now" follows the clock — from 5 AM to noon it leans toward big tasks (energy High, or an estimate of an hour or more); at every other hour it leans toward quick ones (energy Low, or an estimate of 15 minutes or less). The lean is small, one step on the importance dial, so anything overdue or due in the next few days stays on top. A task with no energy level and no estimate is not moved by the clock in the free ranking, though the AI can still judge it from its wording. The AI ranking is told your local date, time and time zone. When the morning ends, or at 5 AM, an AI pick on the card gives way to the free ranking until you press Re-rank, because the card never spends by itself.
- Daily check-in — the header's ⋯ More → Daily check-in now starts the Plan my day session (the same one as Plan my day with AI on Today), which runs on your own AI. The in-panel check-in described here is what the scheduled daily check-in setting produces, opened from its notification: an interactive conversation that acknowledges recently completed tasks (today and yesterday), reviews what's due and slipping, and proposes today's focus — at most 3 tasks, even when your planning profile says you like more in a day, so Plan my day is the way to get a longer list. Starting it runs one bounded AI call (a single click, no confirm pop-up); then you can reply (each reply is one bounded AI call) and Accept the proposed focus. The opening turn and every reply stream in as they are written, and the thread follows the newest line while you are at the bottom of it — scroll up mid-answer and it stays where you put it until you send a message yourself — the answer paints token by token in a bubble (a three-dot bounce shows only until the first token arrives) rather than leaving you on a spinner for the whole call. Today's thread is saved, so reopening it doesn't re-spend. Each reply gets that summary of your list as it is at that moment, and that summary is the last thing the AI reads — sent once, in the same message as your question, after the earlier turns rather than before them, so the current list wins over anything the thread said earlier. The earlier turns still come along for continuity — the most recent 12 messages of the thread only, so in a longer check-in anything said before those is no longer sent, and anything from the start that still matters is worth repeating — under a line telling the AI they were written against the list as it was then: a task it mentioned this morning that you have since finished, renamed or removed is treated as a change to your list, not as a mistake it apologizes for, and it is never re-suggested (
buildCheckinTurnMessagesintasks-v2-checkin-prompt.ts). It can also run automatically once a day at an hour you set — that scheduled run is off by default (opt-in). If the scheduled run's AI call fails (an outage or an unreadable reply), the check-in opens on the plain local summary, the same "The AI was unavailable" notice shows above the thread, and a later 15-minute tick tries the AI opening again — only while you have not replied or accepted, and at most three failed openings a day; a failure after the paid reply came back ends that day's scheduled attempts so the same call is not paid for again. The hour is read on the computer's own clock, so the check-in follows the time zone the machine is in rather than a separate setting: travel with a laptop and the check-in moves with it. There is no longer a time-zone picker — the retiredtasksV2CheckinTimezonesetting is deleted on upgrade, andtasksV2CheckinHouralone decides when the run fires. When more tasks are overdue than the preferred number of items a day saved in your planning profile, it suggests replanning and names the tasks that could be dropped, rescheduled, or delegated.- Long replies are finished, not cut short. The check-in asks the AI to keep each reply short, with the most important items first and an offer to go through the rest, and the opening briefing has the same room to write as a follow-up reply. If a reply or the opening is still cut off, Tasks asks the AI once more to continue exactly where it stopped and joins the two parts before showing anything. Only if that second try is also cut off do you see the partial text with "This reply was cut off before it finished." and a Continue button, which sends "Please continue where you left off." A cut-off reply never shows "couldn't respond". Each cut-off is counted for quality tracking (
tasks_v2 / ai_reply_cut_off). - Start today's check-in over. When today's check-in has messages, a Start today's check-in over button (disabled while the AI is answering) asks you to confirm, then clears today's check-in conversation, suggested focus and accepted mark and opens a fresh check-in. Any Plan my day plan for the day is kept. Clearing is free; the fresh check-in that follows is a normal paid open. The CLI route is
POST /tasks-v2/checkin/restart. - A reply written against yesterday's check-in is refused, never moved to today's. Each reply names the check-in it belongs to. Send one after local midnight while yesterday's thread is still the open one, and Tasks refuses it with "This request was for
<date>, which has passed. Ask again." instead of quietly opening today's check-in and dropping your words into a conversation you have never seen. Accept and Start today's check-in over already answered this way; send now matches them. A reply naming a check-in that no longer exists says "That check-in could not be found." A caller that names no check-in keeps the older behaviour, so the sends that carry no id are unchanged. - Yesterday's recap is frozen. The morning line "Yesterday: completed N of M planned (K carried over)" uses the numbers frozen at close-out, not live task data, so it does not change if you edit tasks afterwards.
- Long replies are finished, not cut short. The check-in asks the AI to keep each reply short, with the most important items first and an offer to go through the rest, and the opening briefing has the same room to write as a follow-up reply. If a reply or the opening is still cut off, Tasks asks the AI once more to continue exactly where it stopped and joins the two parts before showing anything. Only if that second try is also cut off do you see the partial text with "This reply was cut off before it finished." and a Continue button, which sends "Please continue where you left off." A cut-off reply never shows "couldn't respond". Each cut-off is counted for quality tracking (
Who runs it: every Tasks AI call runs first on your own Claude account, with the default Claude model you chose. Only when your Claude account cannot answer (you are not signed in, the sign-in expired, your plan's usage limit is used up, or Claude is unavailable) is the call paid for in the order you set under Settings → My API Keys — your own OpenRouter or Claude key first, then Omniscio credits, then the included allowance (see Built-in AI helpers). OpenAI and Gemini keys are not used for Tasks. Cost controls on those fallback calls: a cheap model — on OpenRouter (your key or the allowance) it is Haiku 4.5, and with your own Claude key your task text and session excerpts go to Anthropic. There is no Tasks daily budget: no Tasks AI call is ever refused, cut back or carded for what was already spent. Every call is cost-tracked, and each feature has its own switch to turn its AI off (it falls back to the free local ranking). AI arrange (used by the Today plate's auto-arrange) runs on the Model for arranging your day setting, else your default Claude model, else Sonnet; when the company lane pays for it (you hold no key of your own, or your key order puts the company lane first) it uses the strongest Claude model that lane offers if your pick is not one of them. Its own switch turns it off; when it is off or the AI fails, the surface degrades to the free deterministic arrange.
A command-line approval for "today" is good for that day only. The check-in run, the check-in message, the Plan My Day open and the Plan-my-day agent launch each pin the plate day into the queued request when it is made, and the approval card names that day (for example "Run the task check-in for Mon, Oct 19 (2026-10-19) with AI") instead of saying "today". When the request is approved after that day has ended, nothing runs and nothing is billed: it is refused with "This request was for 2026-10-19, which has passed. Ask again." A Plan-my-day agent request for a later day stays approvable until that day passes. A day that has already passed is refused up front: POST /tasks-v2/arrange, /tasks-v2/arrange-ai and /tasks-v2/launch-plate-agent answer a past plateDate with 409 before anything is arranged or any approval card is queued, with the message "<date> has already passed, so it cannot be arranged. Arrange today (<today>) instead by sending plateDate <today> or leaving it out." and suggestedPlateDate: <today> in the body. Arranging today already looks at every open task, including Inbox items and overdue ones. An approval card otherwise lasts a week, which is far longer than the day these requests are about.
The scheduled check-in's switch and hour are rows on the Tasks settings page (search Settings for "check-in"); the on-demand check-in works whatever they say.
M2 "Right now" status: The pick lives on the Today card described above, not on a header pill — by user request the header carries none — and the card reads only the stored pick, so nothing spends without a click.
Plan my day (M6)
Status: in development — part of Tasks, on by default once Tasks is on. Switch it off or back on with the Plan My Day row in Settings → Tasks · Caught (under More Tasks settings; search Settings for "Plan My Day"); while it is off, the Show day-load labels in Plan My Day row in the AI planning group is grayed out. If it has been switched off, opening it from Tasks also asks "Turn on Plan my day?" with a one-click Turn on.
What it does
Plan my day opens a chat. Every Plan my day button — Plan my day with AI on Today, the empty-day Plan my day card, Plan my day in a session in the All Projects Today band, the Plan My Day card on the Sessions start screen, and ⋯ More → Plan my day or ⋯ More → Daily check-in — starts the same Plan my day session (the plate agent in part 6), then switches the Tasks panel to its Sessions tab with that chat open. The session reads your Today plate, your open board and your priorities, talks the day through with you, and captures what to focus on once you approve. It runs on your selected AI model and uses AI on every turn. While Plan my day is switched off, the button first asks "Turn on Plan my day?".
The single-call review panel described below is still in the app, but no button in the app opens it now. Its engine still answers the command line (POST /tasks-v2/plan/open drafts a plan, POST /tasks-v2/plan/commit saves one), so the rest of this section describes that engine. It goes beyond reviewing what's due: instead of a multi-turn conversation, it makes a single AI call, shows you a proposed plan — a focused task list, any new tasks it suggests adding, and optional "next action" splits for big items — and asks you to review and accept. Once you accept, the tasks are created and your focus list for the day is set.
The review panel has three states. Idle: no plan today yet, and a cost-gated Start button opens the AI call. Reviewing: the AI's proposal is shown in four editable sections — "Today" (existing task focus), "Break down" (splits), "New to capture", and "About you" (the changes the AI proposes to what it knows about you).
The About-you section renders only when the AI proposes a profile change, and it is a full member of the review, not a silent side effect: every proposed field is a keep/remove row, and the commit carries only the kept subset — a field the user drops is omitted from the delta entirely rather than sent as null, so a dropped field never blanks a value the user typed. Committed: the plan has been accepted and a calm confirmation is shown; it also says planning runs once a morning and points at Today's Re-arrange and AI arrange for changing the day, because a second Plan my day run the same day is a no-op by design and reopening the panel would otherwise read as a dead end. Accept plan sits in a fixed footer under the scrolling review list, so it never scrolls out of view, and a failed save shows its error there too. Re-plan sits beside it: a review you have not accepted yet goes stale as the day fills up, so Re-plan asks for a fresh proposal from your current tasks and the preferences you have saved since — it replaces the proposal rather than adding to it, and costs another AI call. On a starter (fallback) plan, Try AI again shows a spinner while the new AI call runs.
How the single-call plan is used (command line):
POST /tasks-v2/plan/open(an approval card first, because it is a paid AI call) drafts the plan.- The AI returns its proposal: a message, a suggested focus list, any new tasks, and any breakdowns.
- The caller reviews the proposal and may trim the focus list and new tasks.
POST /tasks-v2/plan/commitsaves the plan. New tasks land in your Inbox; split next-actions are created as children of the parent task; Today's plate is arranged automatically from the accepted focus, in the plan's order: in the manual sort the plan's tasks come first, in the order the plan listed them, and anything else already on today follows in its usual order (same as accepting a daily check-in — you shouldn't need to click "Re-arrange today").
If you already accepted today's check-in, drafting a plan keeps the focus you accepted as the plan's starting list instead of replacing it with the AI's suggestion, so the check-in never shows "Accepted" over a list you did not choose.
Once today's plan is accepted — from the morning check-in or from a drafted plan — the review panel shows Today's plan is set with the accepted focus tasks listed, and a Plan again button. Plan again first asks "Make a new plan? Your current plan stays until you accept the new one."; the new proposal then shows "Your current plan stays until you accept this one." and only accepting it replaces today's focus and re-arranges Today (tasks-v2:plan-open with { again: true }; the command-line open does not take it).
A plan accepted after midnight becomes the new day's plan. A proposal made yesterday that you had not accepted stays on screen after midnight; accepting it commits it as today's plan, arranges today's plate, and shows a small "Planned for <weekday>" note. It never overwrites a day that already has an accepted plan ("Today already has a plan. Open Plan my day to see it."), and a proposal two or more days old is still refused.
Coming back after time away. When the last day you accepted a plan was before yesterday, the plan says how many things landed in your Inbox while you were out (for example "4 things landed in your Inbox while you were out.") and adds those open Inbox tasks to the focus, each with the reason "Landed in your Inbox while you were out", up to the focus limit; when not all of them fit it adds "N of them are in the plan for today." Only top-level, open, actionable Inbox tasks created since the day after your last planned day count (tasks-v2-plan-away-inbox.ts).
The AI personalizes the plan from your planning profile — how many items you like to focus on, your standing priorities, and any notes you've added about yourself. That profile also changes as you use it. Each accepted plan updates how many tasks you like per day. When the AI thinks it has learned something else about how you plan — your preferred task size, standing priorities, notes about yourself, work hours, meeting buffer, task-type preferences or energy patterns — the review lists it under About you (proposed changes). Remove any you do not want before you press Accept. The ones you keep never replace words you wrote: for the text fields, what the AI learned is added on a new line below your text (skipped if it is already there or would make the field too long), and an empty field takes the AI's text as is. The two number fields (items per day and meeting buffer) take the reviewed value, and the buffer is accepted only within 0-240 minutes (MEETING_BUFFER_MINUTES_MAX, up to 4 hours), so a garbled reply cannot make every future day unschedulable.
To check or change the profile afterwards, open Tasks → ⋯ More → About you, edit any field and press Save. Clear there erases the whole profile.
The same count sizes AI arrange on Today: it picks up to your preferred number of items minus the tasks you put on Today yourself (so it adds nothing when your own picks already fill that number), aims for 5–7 when there is no preference yet (and never puts more than 7 on Today in that case, even if the AI suggests more), is told the current hour so an afternoon arrange plans only the rest of the day, and reads only your first 200 open tasks. Tasks you pinned stay put; an earlier AI pick it does not choose again comes off Today but stays in its project.
Your planning profile is separate from the short notes Tasks writes about how you work. You see both under ⋯ More → About you. Reset learned memory erases the notes Tasks wrote and the day-by-day history they come from. It does not change your planning profile. Use Clear for that. Notes you wrote or edited yourself are kept. A reset can't be undone. A note you dismiss is hidden right away. Notes you reset or dismiss are removed for good by the daily cleanup 30 days later. That 30 days is fixed: it does not follow your Data Retention setting, and turning off the deleted-task cleanup switch does not keep them.
Clear planning conversations (also under ⋯ More → About you) permanently deletes every daily check-in conversation and every Break-it-down interview for the account you are signed in to, after you confirm. Nothing is kept to restore, so undoing a task delete later never brings an interview back, and an older backup cannot put them back either. Your tasks and learned notes are not touched. The button is not shown on a paired phone. The same action is a CLI route, POST /tasks-v2/ai-conversations/clear, which needs the full-trust token and takes no body.
Erase all local data also removes the check-ins, interviews, Plan chat, learned notes, day-by-day history and planning profile, and every "Export as JSON" folder on the computer. Your tasks and lists are kept, and the erase report names both as kept on purpose. Each list's context text is cleared, though, and the tasks-v2.md file Tasks keeps for your agents is removed (Tasks writes it again the next time it is on and something changes). Restoring a backup made before the erase brings the list context back. An email address, phone number or key you type into a check-in or an interview is masked before it goes to the AI provider; the thread still shows exactly what you typed.
Your AI Coaching "About you" profile is a separate thing. By default it is not sent to the planner: the shared Tasks AI lane is never treated as first-party Anthropic for your coaching profile, even when your own Claude key pays for the call, and the profile only ever goes to Anthropic's own AI unless you agree. The first time it would help, Plan my day asks once: "Use your coaching profile to plan your day?" Answer Yes, use it and your coaching profile is sent from then on for Plan my day and the daily check-in to OpenRouter, the outside AI service the Tasks AI runs on (which stays the case even when your own Claude key pays for the call); Not now sends nothing and does not ask again. Take a Yes back any time in Settings → Mission Control → About You.
Cost: one AI call per plan (the morning's first, plus any Re-plan you ask for), run on your Claude account first, then your saved keys or credits; no daily budget limits it. If the AI is off, nothing is charged and a free local plan is shown instead. If the request starts but times out, or the reply cannot be read, the free local plan is shown too, but a fallback request on a key or credits is still recorded as spent — a timed-out request counts at the most it could have cost.
The arrange pass uses the Model for arranging your day override, or your default Claude model on the Tasks settings page, right under the AI-arrange switch (Settings search "arrange model"); non-Claude models are skipped and Sonnet 5 is the final fallback. On the company lane, a model the lane does not carry is swapped for the strongest Claude model the lane offers (your own OpenRouter key keeps your pick), and the arrange call is recorded at the price of the model that actually ran.
Related
The overview, the other parts, and everything else worth reading next all sit on Tasks (markdown outliner, in development).
Last verified 2026-11-03