Sakura

SAKURA

Sign in with Google to sync your documents, settings, and templates across devices. Fully optional — everything keeps working locally if you skip this.

Help

New here? Take the guided tour.

Quick reference for everything below.

Curious why an outliner helps?

See it all come together, end to end.

Pick a category on the left for reference on any part of Sakura — shortcuts, panels, AI, and everything else.

Core editing

Enter / F2 → edit
Enter → new sibling below (while editing)
Shift + Enter → split at the cursor — text after the cursor moves into a new sibling below, text before it stays put
Ctrl/Cmd + Enter → new child — works whether you're editing a node or just have one selected
Type → start editing

Navigation

Click → select
Shift + Click → range
Ctrl/Cmd + Click → multi-select
Ctrl/Cmd + A → select all

Quick Assist

Lives in the header, where a plain search box used to sit. Ctrl/Cmd + K focuses it and opens the results dropdown from anywhere in the app; clicking directly into the box does the same. Type a plain-English instruction and it's carried out immediately: "get me toolbar," "I don't want file explorer," "get rid of pad," "toggle dark mode" — or just type anything else to search documents, notes, tags, settings, and help. Not to be confused with Quick Insert (Ctrl/Cmd + Space) — a much smaller character-insert menu that opens instead whenever you're actively editing text, described below
An empty box shows a "Try things like…" tip with a handful of example phrases — click any of them to run it as a search immediately, same as typing it yourself; nothing executes without a second click on the actual result. One of the examples, "apply editor's choice," is the fastest way to reach the preset covered below without needing to open Settings at all. Turn the tip off in Settings → Bars & Menus → Quick Assist if you'd rather not see it — the box itself works identically either way
To search just one category instead of all eighteen at once, type a category prefix before your query — "notes: budget" only searches Notes, "settings: dark" only searches Settings. Short aliases work too (doc/docs, note/notes, tag/tags, feature/features, and so on). Pad's three sub-tabs each have their own prefix too — "qa: rollout" searches only Q&A questions/answers, "diagram: flow" only Diagrams (title + note), "remark: EDI" only Remarks (text + person + context) — separate from plain "pad:", which still means just the Notepad tab's free text. Don't want to remember the exact word? Click the icon at the left of the box, or press Space while it's completely empty — either shows a picker inline in the dropdown, navigable the same way search results are (↓ / ↑ to move, Enter to pick). Above the categories, the same picker also lists Show, Hide, Toggle, and Run — picking one inserts that bare word instead of a category prefix, which does something different: rather than scoping a search, it browses everything of that kind (see the bare-verb bullet below). Either way, choosing an entry leaves the dropdown open, cursor ready for you to keep typing and narrow further. A category prefix with nothing typed after it (just "notes: " on its own) shows no results rather than browsing the whole category — the box doesn't have a "show me everything in here" mode for categories, only "search within here," so you still need something to search for after a category prefix specifically; the four verb words above are the exception; each of those does browse everything on its own with nothing else typed
Typing just a bare word — "show," "hide," "toggle," or "run," nothing else — lists everything of that kind instead of requiring you to already know a specific setting's name: "show" lists everything currently hidden that could be shown, "hide" lists everything currently shown that could be hidden, "toggle" lists everything either way regardless of its current state, and "run" lists every available Run action ("new document," "duplicate node," "insert decision log," "editor's choice preset," "documentation mode preset," and the AI actions covered below). Keep typing after the bare word to narrow the list further by name — "show tool" narrows to just toolbar-related results, the same as typing "show toolbar" would, just built up in two steps instead of one
On a Mac, if the system's own Spotlight is bound to Cmd+Space (its default), it takes priority over the browser and this shortcut won't fire — either change Spotlight's shortcut in macOS System Settings → Keyboard, or remap this one from Settings → Keyboard Shortcuts
Toggles and search are rule-based, not AI — a fixed set of recognized phrases (show/get/enable/turn on… vs. hide/remove/get rid of/turn off…) matched against a fixed list of settings. Nothing is sent anywhere for these, no API key needed, and it never improvises — if it doesn't recognize a phrase, it says so rather than guessing. The AI Run actions below are the one exception: those do call your configured AI provider, the same as clicking their toolbar/right-click equivalent would
Covers the everyday visibility toggles: toolbar (and its group labels), file explorer (and its filter box), Pad, zen mode, dark/light mode, auto-rewrite on commit, tree lines, timestamps on hover, spellcheck, backlinks, compact rows, word wrap, outline numbering, the Quick Assist search box itself and its icon-only mode, icon-only Import/Export in the app bar, the hover toolbar, delete confirmation, decision log colors, accent-colored branches, expanded toolbar, and the startup tab/blank-session preferences — plus every on/off switch for a whole capability, whether it lives in Settings → Features (Note Panel, Code Block, Pad Panel, To-Dos, Meetings, Journal, Recap, Library, AI Capabilities, Tagging, Quick Assist, Quick Insert) or inside that panel's own section (Decision Log, Focus, Backlinks, Hover Toolbar). It does not cover destructive actions (reset, delete, empty trash), Pad's individual tab toggles (Q&A/Diagrams/Mind Map — those stay Settings-only for now), or multi-value settings (themes, sliders) — those still live in Settings, on purpose
Note Panel, Code Block, Tagging, and Decision Log are reachable with slightly longer phrases than you might expect — "note panel feature," "code block feature," "tags feature," "decision log feature" — rather than their bare names. That's deliberate: those bare words are also what Quick Assist searches for as content (or, for Decision Log, what every inserted entry starts with), so a short keyword there would rank a toggle above the search hit you were actually looking for. Pad Panel is qualified too ("pad feature") for a different reason — bare "pad" is already claimed by the command that opens or closes the Pad panel for this session, which is a separate thing from turning the whole feature on or off
Below command matches, a separate row type labeled Run covers one-off actions instead of toggles: "new document," "duplicate node," "duplicate," "editor's choice preset" (or "apply editor's choice"), and "documentation mode preset" (or "apply documentation mode") — or type the bare word "run" with nothing else to list all of them at once, the same way a bare "show" or "hide" lists every matching toggle. These don't have an on/off state to show or an existing value to report — they just do the thing once, the same as clicking the equivalent toolbar or Settings button would. Only safe, additive, undoable operations are ever reachable this way; nothing destructive is offered as a Run action, same rule as the toggles above. "Duplicate node" needs a node selected first — if nothing's selected, it still shows up (so you know it exists) but greyed out with "select a node first" instead of a working button, rather than either hiding it or letting a click do nothing unexplained. Undo means different things per action: for "new document" and "duplicate node" it's the regular document Undo; "editor's choice preset" instead restores every setting it touched (toolbar, file explorer, Pad, hover toolbar, accent, text color, right-click menu) back to what they were immediately before, since none of those live on the document undo stack
AI Run actions — also listed under "run," these are the exception to "rule-based, not AI" above: "rewrite document" (the whole open document, same as right-click → Rewrite document), "rewrite this node" / "rewrite selected" (current selection, single or multiple nodes), "expand node" / "expand into subtree" (needs exactly one node selected), "summarise selection" (needs 2 or more nodes selected), "suggest tags" (needs exactly one node selected), "generate outline" (opens the same topic-prompt dialog as its toolbar button), and "restructure text" (opens the same paste-text dialog). Each needs an API key configured in Settings → AI → Provider and disappears entirely if AI Capabilities (Settings → Features) is off — same requirement as the toolbar buttons they reuse. Unlike the rule-based actions above, these take a moment to run and can fail (expired key, provider rate limit) exactly like clicking their toolbar equivalent would — a missing key or a failed call surfaces the same error dialog either way, not a silent no-op. Undo works the regular way once a result lands, since each reuses the exact same function — and therefore the exact same pushUndo() call — as its toolbar/right-click counterpart
Below any command and action matches, the same box also surfaces matching documents, node text, notes, tags, folders, templates, and each document's Pad — including its Q&A, Diagrams, and Remarks sub-tabs individually, not just the Notepad text — plus To-Dos, Meeting Notes, Journal, Library, and settings/help topics. These rows are labeled Go to instead of Show/Hide, since picking one navigates you there rather than changing anything, and each shows a short snippet of where the match sits so you're not guessing what you're about to open
Clicking a "Go to" result does the right thing for its type: a node-text match jumps to that node and highlights the match; a Notes match opens that node's Note panel; a Tags match jumps to the tagged node; a Pad, Q&A, Diagrams, or Remarks match opens that document's Pad to the right sub-tab (a Diagrams match opens that specific diagram); a To-Dos/Meetings/Journal/Library match opens the Hub to that entry
Skips Trash by default. Settings → Trash → "Include Trash in search" surfaces matches there too, across every content type above, clearly marked "In Trash"; clicking one reveals it in the Trash section rather than opening it directly, since restoring is a deliberate choice
Rich-text fields (Notes, Pad) are stripped to plain text before matching, so a search never hits raw HTML markup and snippets always show readable text
The file explorer's own filter box only matches document/template names, not their contents — for anything inside a document, this box is the one that actually looks
Every document has a status chip — No status by default, or Draft, Review, Approved, Rejected — next to the title field. Click it to open a small menu and pick one; it's a menu rather than a cycle-through-in-order control like the diagram/decision-log status chips, since jumping straight to the one you want beats clicking past the others first. Existing documents start with no status until you set one; it's never assumed. Duplicating a document resets its copy to No status rather than carrying the original's over, same reasoning as why the title gets "(copy)" appended instead of reused verbatim — a fresh copy hasn't itself been through review or approval. Version History restores status alongside everything else for versions saved after this feature shipped; older versions predating it simply leave the current status untouched on restore rather than guessing
Settings → Bars & Menus → Quick Assist controls the search side of this specifically: a switch to turn search results off entirely (command-only), how many results show at once, and which categories (Documents, Notes, Tags, Settings, Help, and so on) are eligible to appear at all
As you type, the list runs in a fixed order top to bottom — matching commands first, then Run actions, then search results — each labeled with what it'll do ("Hide File Explorer," "Show Toolbar," "Run: New document," "Go to: …") and, for toggles, the setting's current state; ↓ / ↑ moves through the combined list (disabled Run rows are skipped, since there's nothing to activate), Enter runs or opens whichever is highlighted, Esc closes and clears the box, or click any of them directly
Every command and Run action shows a toast with an Undo button — nothing is ever silently applied. Just typing a bare toggle name with no verb ("toolbar," "file explorer") flips its current state rather than doing nothing. If a Run action has its own confirmation step (Insert decision log asks who's deciding first) and you cancel that step, nothing happens and the toast says so instead of falsely claiming success. "Go to" rows are navigation, not a change to anything, so there's nothing to undo for those
If a phrase matches more than one setting, all of them show up as separate ranked options instead of Quick Assist silently picking one for you

Quick Insert

Ctrl/Cmd + Space while actively editing any text — a node, the Pad, a Note, a Code block, a To-Do, a Meeting Notes field, anywhere — opens a small character-insert menu. Same shortcut, same menu, everywhere you're editing text; the full Quick Assist is a separate shortcut (Ctrl/Cmd + K) and stays free even mid-edit, so the two are never in competition for the same key
Actions, in order: Insert em dash, Insert en dash, Insert arrow, Insert check mark, Insert cross mark, Insert middle dot, Insert date/time. Which of these appear (and their fixed order) is configured in Settings → Editing → Quick Insert → Menu items
No typeahead — unlike @mentions or the full Quick Assist, this menu doesn't filter as you type. ↓ / ↑ moves through it and wraps at either end, Enter runs the highlighted action, Esc or resuming normal typing dismisses it. There's deliberately nowhere for a typed query to live except the text you're editing, which this avoids touching
Settings → Editing → Quick Insert → Icon-only row switches the menu from a vertical icon+label list to a compact wrapping row of icons — same actions, same order, same keyboard navigation, just a smaller footprint. Hover an icon for its name
Every action splices directly at the cursor and keeps you in edit mode, so you don't lose your place — the fastest way to get these in, and the reason the menu exists
Node-only actions — Note, Tags, Add question, Add remark, Rewrite node, Version history — live in a node's right-click menu instead, not here, since Quick Insert is deliberately the same small menu everywhere rather than growing extra items only inside the outline editor
Version history… (right-click a node → in the "notes" group of the More panel, or promotable to the quick row via Settings → Right-click menu → Menu items) — shows earlier text for just this one node, drawn from the document's regular version history (captured automatically roughly every 10 minutes of active changes — see the Version History entry below) — skips straight past versions where this particular node didn't change, so a document with 20 saved versions might show far fewer than 20 rows here. "Restore text" replaces only this node's text, through the normal undo stack (Ctrl/Cmd+Z undoes it like any other edit) — nothing else about the node (note, code, tags, decision log) is touched
Settings → Editing → Quick Insert turns the whole thing off — Ctrl/Cmd+Space then does nothing while editing text, rather than falling back to anything else

Appearance & Display

Theme & accent — Settings → Appearance: Light/Dark theme, seven accent colours, and an intensity slider. Auto theme sits below it — Off (the Light/Dark switch only, changed by hand), System (follows your OS/browser's dark mode setting live, no polling needed), or Schedule (switches at hours you set, checked once a minute plus whenever the tab regains focus, to catch changes that happened while it was backgrounded or the machine was asleep). Manually clicking Light/Dark in System or Schedule mode holds as a temporary override until the schedule/system value naturally catches up and agrees with it again — e.g. forcing dark during a scheduled light window holds through the next scheduled dark transition and then continues following the schedule normally from there, rather than reverting within the next check. "Accent colour on branches" (Settings → Layout) tints the tree connector lines with the accent too; off by default so branches stay neutral
Content font (Settings → Appearance, below Theme) — Sans-serif (Inter) or Monospace (Fira Code) for the text you actually write: the outline, To-Dos, Pad, and Note all switch together, so it's one consistent choice rather than a per-panel setting. Tree-line connectors (│├└─), code blocks, and inline code always stay monospace regardless of which is picked — those are structure/code, not prose, and switching them would hurt rather than help. Sans-serif is the default; Monospace is the original look for anyone who prefers it
Chrome background (Settings → Appearance, below Accent) — five presets: Default (follows the Light/Dark theme), Slate, Sand, Ink, and Rose. Recolors the toolbar, file explorer, status bar, app bar, settings, help, and menus; the editor/canvas writing surface is untouched, so your outline keeps reading against its own background regardless of which preset is picked. Independent of both the accent color and the Light/Dark theme — any combination of the three is valid
Text color (Settings → Appearance, below Chrome background) — four presets for node text specifically: Default, Black, Charcoal, and Slate. Separate from the accent color (which colors buttons, borders, and highlights, not node text) and from Chrome background (which colors app chrome, not the outline itself)
Status bar items (Settings → Bars & Menus, under the Status bar switch) — which individual pieces show in the bottom strip: node count, max depth, word count, current depth, current node's word count, saved status, backup connection status, the auto-rewrite indicator, theme chip, text size control, and author field, each independently hideable. The master Status bar switch above still controls the whole strip; this is finer-grained, for keeping only what you actually use. The "⚠ Back up your data" reminder chip alone isn't in this list on purpose — it's an actionable alert that only appears when your backup is genuinely overdue, and permanently hiding an alert defeats it; if it fires too often, raise the reminder interval in Settings → Data & backup instead
Status bar item order (Settings → Bars & Menus, right below Status bar items) — drag to reorder the right-hand cluster: saved status, backup connection status, the backup-overdue reminder, the auto-rewrite indicator, theme chip, text size control, author field, and Quick Assist. Hidden items keep their place in this list, so turning one back on doesn't reset where it sits. The left-hand counters (node count, depth, word count) are a separate, fixed-order cluster and aren't part of this — only the right side is reorderable
App bar items (Settings → Bars & Menus, under the App bar switch) — hide Help, Print, Import, and Export individually if you rarely reach for them, to leave more room for Documents/Templates. Settings and Documents/Templates always stay in the app bar — Settings needs to stay reachable (same reasoning as the App bar master switch above), and Documents/Templates is core navigation. Just below that, Icon-only Import/Export collapses Import and Export down to their icons instead of hiding either outright — a middle ground when you still want one-click access but not the label text; hover either icon for its name. Both are separate from, and finer-grained than, "Declutter app bar" (in the app bar's own right-click menu), which hides the title field, file explorer toggle, and Documents/Templates buttons instead — all of these can be combined
Toolbar — Settings → Bars & Menus → Toolbar: "Show toolbar" hides/shows the icon bar (also toggled by the floating button next to the editor or Ctrl/Cmd + Shift + X); "Group labels" adds a caption under each icon group; "Word wrap" controls whether long lines wrap in the editor or scroll horizontally instead; "Expanded" shows the Extras group as buttons instead of a menu. "Sections" lets you hide whole icon groups (History, Structure, Move, Format, Fold, Insert, AI, Extras, Delete) for a leaner bar, and "Insert buttons" trims individual buttons within Insert without hiding the whole group
Editor's choice preset — Settings → Presets → Apply, or Quick Assist as a Run action ("editor's choice" — see below, also one of the clickable example phrases when Quick Assist opens empty): one click reconfigures the toolbar, file explorer, Pad, hover toolbar, status bar, app bar, Quick Assist location, editor backgrounds, Quick Insert, layout, and Presenter mode to a curated set for writing and presenting. Toolbar, file explorer, and Pad start hidden; hover toolbar is turned off; tree lines are hidden for a cleaner indented view; the status bar keeps only Words, Depth, Saved status, and the auto-rewrite indicator, hiding node count, max depth, node word count, the backup-connection chip, the theme chip, the text-size control, and the author field (the Secure Storage lock chip is left alone regardless of this preset, since it only shows itself while actually locked and isn't density-related clutter to begin with); the app bar stays visible with Quick Assist docked in it. Your accent color and node text color are left untouched — this preset is about density and visibility, not color. Note, Code block, and Pad all switch "Match editor background" off, so each reads as visually distinct chrome rather than blending into the outline canvas. The toolbar's own button set is curated for whenever you bring it back (Highlight, Text color, Collapse, Expand, Date/time, Decision log, Note, Code block, Tags, Rewrite, and Delete); the right-click quick row gets Focus, Rewrite, Note, Code block, Date/time, Decision log, Tags, and Delete — Focus lands there specifically because this preset turns off the hover toolbar, its usual home. Quick Insert (the Ctrl/Cmd+Space character-insert menu) switches to its icon-only row with every character action enabled. Presenter mode gets the laser pointer and branding turned on, "Open Preview straight into Presenter mode" turned on, and note cards set to expand automatically; the floating Presenter Notes panel's own toolbar is trimmed to Bullet list, Timestamp, and AI Rewrite. Everything not mentioned is hidden; you can still adjust any of this individually afterward in Settings, wherever that particular control normally lives. However you apply it, an Undo appears — Quick Assist's own toast gets one immediately, and Settings shows a persistent Undo button next to Apply that survives a page reload, so closing the toast or navigating away doesn't cost you the ability to revert. That undo restores every one of the settings above to what they were immediately before, in one step. Applying the preset again (from either entry point) replaces whatever undo point existed with a fresh one taken at that moment — it's a single-level undo, not a history
Documentation mode preset — Settings → Presets → Apply (second card, below Editor's choice), or Quick Assist as a Run action ("documentation mode"): the opposite instinct from Editor's choice, built for long-form structured documentation rather than presenting. Toolbar, file explorer, Pad, and status bar are all fully visible with nothing trimmed; hover toolbar is on; tree lines stay visible for navigating deep hierarchies; Quick Insert shows full labels with every action available. Decision Log, Tagging, and Backlinks are all switched on, with Decision Log entries shown with their status colors and bold field labels. Code block keeps "Match editor background" off so code stays visually distinct from the outline; Note and Pad blend in. Presenter mode's auto-behaviors (laser pointer, auto-launch, branding, auto-expanding note cards) are explicitly turned off — this preset isn't about presenting. Has its own separate Undo, independent of Editor's choice's — applying one doesn't clear the other's undo point
Hide tree lines — Settings → Layout → "Hide tree lines" (or Ctrl/Cmd + Shift + L anytime): drops the │├└─ connector characters for a clean indented view, closer to Dynalist or Workflowy. With it on, every row gets a small dot instead of a triangle fold-arrow — a folded parent shows as a ring with a dot inside, hovering either kind reveals −/+; with it off, parent rows keep the classic ▾/▸ arrow instead. Indentation and folding behave identically either way — this only changes how the tree is drawn, not the structure itself. "Depth guide lines" (right below it, off by default) optionally adds a faint vertical line per indent level once the branch connectors are gone, so nesting stays easy to trace
Section markings — Settings → Layout → "Section markings": toggles the coloured accent bar on section-header rows. On by default; independent of Hide tree lines
Row selection style — Settings → Layout → "Row selection style" controls how selected rows are marked: Fill (default) shades the row background, with the primary/anchor row a shade darker than the rest of a multi-select; Outline traces a thin border, with the anchor row's border and a light fill slightly stronger; Left bar shows an accent-coloured bar on the left edge, thicker on the anchor row; Dot marks each selected row with a small dot, larger and solid on the anchor row. Fill and Outline are neutral/near-neutral; Left bar and Dot use the active accent colour
Layout — Settings → Layout: "Compact rows" tightens spacing between separate rows independent of text size, without affecting line spacing within a single wrapped item; "Text size" is a slider from 85% to 140%; "Limit reading width" caps line length and centers the tree instead of spanning the full window, off by default (Preview has always had its own separate version of this, always on); "Branch spacing" sets the horizontal indent per depth level (2-6)
File Explorer — Settings → Bars & Menus → File Explorer: control whether new folders start collapsed, and whether a title-filter box appears above the documents list
All of the above are saved automatically and restored next time you open the app. "Reset to default settings" (Settings → Data) reverts appearance and behaviour preferences to their defaults without touching your documents

Structure

Right-click a node → a small row of four marker icons (Confirmed, Issue, Parked, Follow-up — see the Markers entry below) sits above everything else, always visible and not part of the customizable menu; below that, a row of icon buttons for the most common actions, defaults are Zoom in, Note, Tags, Rewrite, Delete. Click More for move, sort, rewrite document, and everything else. Customise which buttons appear via Settings → Right-click menu — you can also reset to defaults there. Zoom in is also in the toolbar Fold group and via Ctrl/Cmd + .; zoom out via Ctrl/Cmd + , or clicking the same toolbar button again
Markers — five quick marks (✓ Confirmed, ⚠ Issue, ⏸ Parked, 🚩 Follow-up, ✗ N/A) for tracking status on a node, e.g. while reviewing a document. N/A covers something discarded or out of scope — deliberately a different icon from Issue, so "this needs attention" and "this doesn't apply/was decided against" don't collide on the same cross mark. Set from either Presenter mode (click a line, see the Presenter entry above) or the editor's own right-click menu — same icons, same toggle-to-clear behavior, either place. Once set, a small colored glyph shows on that node in both the editor and Preview/Presenter. In the editor, clicking the glyph filters the tree to that marker — showing every matching node, its full subtree, and its ancestors back to the root (dimmed, so they read as "here's where it lives" rather than more matches); a filter bar appears with a × to clear it. Right-click the glyph to clear the marker itself instead. A marker filter and a tag filter can both be active together — matching nodes must satisfy both (tag matching itself stays a plain per-node check, unlike the marker filter's subtree behavior).
Right-clicking empty space — editor canvas, file explorer, status bar, tab bar, or app bar — opens a separate, smaller menu of chrome-level actions, not the node menu above. Navigate it with ↓ / ↑ and Enter, same as the category picker above, or hover with a mouse — both move the same highlight. Editor canvas: switch theme, hide/show toolbar, hide/show tree lines, and Collapse all/Expand all sit in the menu itself; a "More…" item opens a second menu at the same spot for the rest — hide/show file explorer, Pad, status bar, or tab bar, "Show app bar" (only appears while zen mode has it hidden), and Settings. File Explorer: switch theme, hide file explorer, toggle compact rows, Settings. Status bar: switch theme, hide status bar, Settings. Tab bar: switch theme, new document, close current tab, hide tab bar, Settings. App bar: switch theme, hide/show file explorer or toolbar, and "Declutter app bar" — which hides the title field, file explorer toggle, and the Documents/Templates buttons, but deliberately never hides Settings — hiding it would remove your only way back in. (Quick Assist used to share this exemption when it lived in the app bar too; it's since moved to the status bar, so it isn't part of this menu's options at all anymore.) Every hide action here shows an Undo toast, same pattern as Quick Assist's own toggles. If you hide Status bar or Tab bar from their own right-click menu, that specific spot disappears with it — but the editor canvas's own right-click menu keeps the same Status bar/Tab bar toggles, so it works as a fallback, in addition to the plain switches under Settings → Bars & Menus. Declutter app bar has no separate switch by design, since Settings (reachable via the button it deliberately never hides) already reverses it in one click
Backlinks — while editing a node, type @ to mention another node by name; a dropdown filters as you type, pick with mouse or arrow keys + Enter. The reference renders as a clickable link that jumps to the target. Open the Note panel on any node to see which nodes link to it. Deleting a referenced node removes its links from all other nodes automatically
Tags — select a node and click the tag icon (toolbar Insert group) or right-click → Tags to open the tag popover. Existing document tags appear as chips to toggle; type a new name and press Enter to add. Tags render as #chips on the node. Click a chip in the tree to filter to that tag; click again or × to clear. Multi-select: toolbar button applies to all selected nodes
Checkboxes — click the toolbar button (Insert group) to toggle a node as a checkbox. Or type [ ] / [x] at the start of a node while editing — it auto-converts on commit. Click the box to check/uncheck. Parent checkbox nodes with checkbox children show a 2/5 progress badge; checking all children auto-checks the parent. Checked nodes fade. Multi-select supported
Heading shortcut — type # through ###### (1 to 6 of them) followed by a space at the very start of a node, then finish editing — same auto-convert-on-commit as checkboxes above, and sets the node's real heading level (the same field the Heading toolbar button sets), stripping the markdown prefix. Caps at H6, matching the toolbar's own options — #Text with no space, or seven or more #, doesn't convert
⤴ → insert above
＋ → add child
Tab / Shift + Tab → indent / outdent
⧉ → duplicate node / subtree
Alt + ↑ / ↓ → move
Drag a row → drop above/below to reorder, or onto the middle of another row to nest as its child
Right-click a node with children → More → Sort children (A → Z, Z → A, or by depth — shallowest subtree first)
Extras in the toolbar → sort all top-level nodes the same three ways

Tagging

Select a node and click the tag icon (toolbar Insert group) or right-click → Tags to open the tag popover. Existing document tags appear as chips to toggle; type a new name and press Enter to add. Remove a tag from a node by toggling its chip off in the popover
Tags render as #chips directly on the node row. Click a chip in the tree to filter the entire tree to that tag; click the chip again or the × to clear the filter
Multi-select: select several nodes, then click the tag icon — the popover applies to all selected nodes at once. Tags already on all selected nodes appear toggled on; tags on only some appear in an indeterminate state
AI Suggest tags (AI toolbar group) reads the node text and the document's existing tags, then applies relevant ones directly — prefers reusing tags already in the document before inventing new ones

@ Mentions & Backlinks

While editing a node, type @ to mention another node by name; a dropdown filters as you type — pick with mouse or arrow keys + Enter. The reference is saved as [[Node name]] and renders as a clickable link that jumps to the target node
Open the Note panel on any node to see its Backlinks section — every node that @mentions it appears there as a clickable entry. If no backlinks exist the section is hidden
Deleting a referenced node automatically removes its [[mention]] links from all other nodes
Turning the feature off (Settings → Panels → Note Panel → Backlinks) suppresses the @ dropdown and hides the Backlinks section in the note panel. Existing [[mentions]] in node text still render as clickable links regardless — they navigate on click even when the feature is off

Focus (zoom into a branch)

Right-click a node → Zoom in (or Ctrl/Cmd + ., or the zoom icon in the toolbar Fold group) → only that node and its descendants are shown. A breadcrumb bar appears above the editor showing the path back to the root; click any crumb to zoom out to that level
Exit focus with Ctrl/Cmd + ,, by clicking the same toolbar button again, or by clicking the root crumb in the breadcrumb bar
Focus state is saved with the document, the same way fold state is — closing and reopening it (even in a new session) returns to whatever branch was last focused
Turning the feature off (Settings → Editing → Focus) hides the zoom-in toolbar button and right-click action, and exits any active focus session immediately. Ctrl/Cmd + . also stops working while the feature is off

Formatting

Ctrl/Cmd + B → bold
Ctrl/Cmd + I → italic
Ctrl/Cmd + U → underline
Ctrl/Cmd + Shift + S → strike
Ctrl/Cmd + Shift + H → highlight (caret next to it picks a color)
Ctrl/Cmd + Shift + F → text color (caret next to it picks a color)
Ctrl/Cmd + Alt + 1–6 → apply Heading 1–6 to the selected node(s), independent of tree depth — pick from the dropdown next to Text color, or Ctrl/Cmd + Alt + 0 / re-clicking the same level to return to body text. Combines with bold/italic/underline/strike on top if you want further emphasis

Semantics

[Text] → Section — blue label
(Text) → note — muted pill/chip
!Text → alert — red warning
`Text` → code — inline code
SAP Note 123456 → SAP Note 123456 — auto-linked to me.sap.com, opens in a new tab. Also recognizes "OSS Note", case-insensitive. No markup to type — just write it naturally
@Node name → Node name — backlink to another node; type @ while editing to pick from a list. Open the Note panel on the target node to see all backlinks pointing to it
[ ] task or [x] task → auto-converts to a checkbox node on commit
Tag icon in toolbar or right-click → Tags → tag popover (toggle existing tags or add new ones)

AI Rewrite

button in the toolbar → rewrites the selected node; with multiple nodes selected, rewrites all of them in one batch; with text highlighted inside an editing node, rewrites just that selection
Right-click a node → Rewrite node is in the quick icon row (Rewrite selection, if multiple are selected); click More → Rewrite document to rewrite every node in the tree
button next to Rewrite, or Ctrl/Cmd + Shift + O → Generate outline (AI): describe a topic and it inserts an AI-generated nested outline into the current document — into a blank doc, or after the current selection; also available as "Generate with AI" on the blank-document screen. Uses the same provider/model/key as Rewrite (Settings → AI → Provider), but its own fixed instructions, not the customizable rewrite prompt
Restructure text (AI) — Import ▾ → "Restructure text", or Ctrl/Cmd + Shift + R: paste messy or unstructured text (notes, an email, a transcript) and it organizes it into a tree, cleaning up wording as needed (fixing broken line wraps, tightening long sentences) so it reads well as outline nodes — it will not invent facts that are not in your text, but it is not a verbatim copy either. If the pasted text already has clear structure (bullets, consistent indentation, or literal tree-connector characters), it is parsed directly with no AI call at all. Unlike Generate outline, this always opens as a new document rather than inserting into the one you have open, so an import can never accidentally merge into unrelated work — your original pasted text is kept in that new document's Pad so you can check it against the result
Word import — Import ▾ → "Word (.docx)": converts the file's own heading and list structure directly into nodes — no AI involved when the document actually uses heading styles, since that structure is already explicit in the file. If a document has no heading styles at all (just a wall of plain paragraphs), its text is automatically handed to the same AI restructuring used above instead of being imported as one long flat list. Tables import as a node per row with each column as a child node; images are dropped but left as a "[image]" placeholder so nothing vanishes silently. Always opens as a new document, with the extracted text kept in its Pad. Legacy .doc files (pre-2007 binary format) are not supported — only modern .docx
Word export — Export ▾ → "Word (.docx)": a genuine, editable document — real heading styles (so Word's own navigation pane and outline view work), and a Table of Contents (headings and section markers both) with clickable page references. Page numbers start as "1" until Word recalculates them — select all (Ctrl/Cmd+A) then press F9 once after opening to load the real ones (a note to this effect sits right above the Table of Contents); right-click → Update Field works too for just the TOC, or again any time headings change later. Decision Log entries render as bordered/shaded cards with the same accent-colored status and field labels as Preview. Notes carry through with their actual formatting — bold, italic, underline, strike, and links all become real Word formatting rather than flattening to plain text; bullet/numbered lists inside a note get a simple prefix treatment, and a table inside a note flattens to one line per row (cells joined by " | ") rather than a real Word table grid. Images marked "Feature as diagram" (right-click an image in a note) embed as real pictures, same as PowerPoint export, scaled to fit while keeping their aspect ratio; a diagram that can't be read or converted is skipped with the rest of the document still exporting normally. Diagrams created in the Pad's Diagrams tab and linked to a node embed the same way, controlled by Settings → Export & print → "Include Diagrams in exports". Settings → Export & print → "Cover page" (on by default, shared with PDF) adds a title page — document title, a divider rule, author, and a word count/read time/decisions-logged line — before the Table of Contents, which always starts on its own page and is followed by one before the body content begins. Follows Preview/Presenter's own fold state (like PDF and PowerPoint) rather than Settings → Export & print → "Skip folded sections" — whichever sections are currently folded there are what's left out. Settings → Preview → Presenter mode → "Branding" adds the same small corner mark used in PowerPoint export and the Presenter bar, as a running page footer, plus a wordmark above the title on the cover page itself
PowerPoint export — Export ▾ → "PowerPoint (.pptx)": builds a genuine, editable slide deck using the same slide breakdown as Preview's Presenter mode (Settings → Preview → "Presenter slide breaks at" controls this) — a title slide, then one slide per node at that depth, with its entire subtree (not just its direct children) as bullet text, nested a level at a time (•/◦/▪), and any images marked "Feature as diagram" (right-click an image in a note) — or diagrams from the Pad's Diagrams tab linked to that node — embedded as pictures sized to fit, correctly scaled to their real aspect ratio. A node right-clicked and marked "Slide divider" (independent of that depth setting, and independent of section markers) always gets its own standalone, title-only slide at that point in the deck — same divider node, same title-only treatment as Presenter mode's mid-deck dividers, vertically centered so it reads as a chapter break rather than a second cover page. A node's own marker (✓ Confirmed, ⚠ Issue, ⏸ Parked, 🚩 Follow-up, ✗ N/A) shows as a small colored glyph in front of its bullet. Everything opens fully editable in PowerPoint, Keynote, or Google Slides — text boxes, bullets, and pictures are normal native shapes, not a flattened image of the slide. If a slide's bullets don't fit in the space available, the rest spills onto a new slide titled "<Title> (cont'd)" rather than getting clipped — measured against the actual font, not guessed. A node's Note text rides along with its bullet too, shown with a icon in italic muted text so it reads as an annotation rather than more outline content; a note's own embedded diagram images still become pictures as above, only for the slide's own top node, not deeper descendants. Two more sections get their own slides before the very end, each paginated the same way: a Notepad section (lists flatten to plain paragraphs, unlike Word's version, since pptxgenjs text runs don't have a real list structure to lean on without a lot more plumbing — but a table stays a real, editable PowerPoint table, wherever it was typed relative to the surrounding notes) and a Q&A section (question bold, answer below it, unanswered ones marked "No answer provided") — both follow the same Settings → Export & print toggles Word uses for Notepad and Q&A. An optional closing slide (Settings → Preview → "Closing slide" — "Thank you" by default, or your own text and subtitle) is always the genuine last slide in the deck, after Notepad and Q&A, same as Presenter mode's own closing slide. A table in the Notepad doesn't need to be linked to a node to export as a real table — right-clicking it and marking it "Feature under selected node" (see the Notepad section above) instead moves it to that node's own slide — or, if also marked "Feature as chart", as a real chart instead (Bar, Line, or Pie — whichever type was picked in the Notepad or by right-clicking the chart in Preview/Presenter), stacked below any bullets/images/decision cards already on the slide. A node-linked table only shows up once, on that node's slide — it's left out of the general Notepad slides so the same table doesn't appear twice in the deck (Word and PDF follow the same rule). Whichever sections are currently folded in Preview/Presenter (not the main outline editor, and not Settings → Export & print → "Skip folded sections") are what's left out of the deck — fold or expand a section there first if you want to change what exports. Settings → Preview → "Branding" (off by default) adds a small mark to the bottom-right corner of every slide — "SAKURA" by default, or your own company/team name if set in "Branding text" right below it — the same mark and setting used in the Presenter bar, and in PDF and Word exports too
Sakura Document — Export ▾ → Share → "Sakura Document" (.sakura.json): the one full-fidelity export — everything the document holds, including notes, Decision Logs, Diagrams as real editable draw.io XML (not just a picture of it), Q&A, and Mind Maps. Every other export format is lossy on purpose (OPML/Markdown/plain text keep structure and text only; Word/PDF/PowerPoint flatten into a document no longer editable in Sakura) — this one exists to hand a complete, still-editable document to a colleague also running Sakura. Import ▾ → Sakura Document brings a received file in as a brand-new document; it never touches any other document, folder, or setting already in your workspace — for moving everything in the app at once (all documents, folders, templates), use Settings → Data → Export/Import instead, though note that importing there replaces the entire app's contents rather than adding to it
button next to Outline → Suggest icon: picks a fitting emoji for the selected node's label and adds it as a prefix (e.g. "🛒 Webshop"). With multiple nodes selected, applies one straight away to each, picked from a keyword list first and an exact match against icons already used elsewhere in your documents/templates, falling to AI only when neither matches. For a single node, if more than one good option turns up, a small popover appears above it to pick from; right-click → More → Suggest icons for all nodes runs it over the whole tree (asks to confirm first). Since the icon is just leading text, removing or changing it is normal editing — click the node and edit it like any other text
Batches larger than 30 nodes are sent in sequential chunks automatically, with a status toast showing progress (e.g. "Rewriting nodes 1–30…")
Expand node (AI) — select a single node and click the expand button in the AI toolbar group to break it into child nodes. Useful for dense labels like "Distribution model (OPML, localStorage, multi-tab sync)" → a proper subtree. AI generates 3–8 children; result is undoable
Suggest tags (AI) — select a node and click the tag-suggest button; AI reads the node text and the document's existing tags, then applies relevant ones directly. Prefers reusing tags already in the document before inventing new ones
Summarise selection into parent (AI) — select 2 or more nodes (Shift-click or Ctrl/Cmd-click), then click the summarise button; AI writes a concise parent label and inserts it above the selection, with the selected nodes indented underneath. The bottom-up outlining pattern — build the detail first, name the group after
Settings → AI → Provider → choose a provider from the dropdown, paste the API key for it, then head to the Rewrite section to optionally customise the rewrite prompt. All AI features — Rewrite, Expand, Tag suggestion, Summarise, Generate outline, Suggest icon — share the same active provider and key. "Reset to default" next to the prompt box restores the original grammar-and-spelling-only wording; "Rephrase" swaps in a different canned prompt that's allowed to restructure sentences and vary wording, not just correct errors — both replace the box's contents and save immediately, the same way, so review before relying on either. There's only one active prompt: switching to Rephrase also changes what Auto-rewrite on commit does, if it's on, since auto-rewrite always uses whichever prompt is currently active. "Reset AI settings" at the bottom restores provider, model, and prompt together and clears every saved key
Auto-rewrite on commit (Settings → AI → Auto-rewrite on commit, off by default) — runs Rewrite automatically on a node as soon as you finish editing it, no manual click needed. Only text typed by hand counts: if any part of an edit session came from a paste or drop (checked via the browser's own paste/drop signal on that edit), that node is skipped for auto-rewrite on that commit — paste it, then use the manual Rewrite button if you want AI on it. Committing a node with no actual text change (e.g. clicking in and back out without editing) is also skipped — it doesn't re-queue nodes that haven't changed. It doesn't send one request per node: committed nodes are queued and sent as a single batched request once either an idle pause passes or enough nodes have queued up, whichever comes first — both the pause length (1-300s, default 120s) and the queue size (2-50, default 5) are configurable in Settings, not fixed — built this way specifically to avoid exhausting provider rate limits during continuous note-taking. A small ✦ dot appears next to any node currently queued (or mid-batch, waiting on the response) — click that dot to pull just that node back out of the queue before it's sent — separate from the status bar chip, which shows the overall off / idle / queued-count-with-a-live-countdown / paused / rewriting state and doubles as the control itself: click it to turn auto-rewrite on or off directly (see below for what clicking does instead while it's paused), right-click it for its settings (exclusions, minimum word count, idle pause, queue size). The countdown ticks down to the batch's send time and resets to the full idle-pause value every time another node joins the queue — a fresh edit means a fresh wait, not a shorter one. Both the queued count and the countdown are independently hideable from the chip (Settings → AI → Auto-rewrite on commit, below the queue-size setting) if either feels like more detail than you want at a glance — the chip still distinguishes "something's queued" from idle even with both turned off, just without the numbers; the tooltip on hover always shows the full detail regardless of what the chip itself displays. Exclusions (checkboxes, decision-log field lines, headings, [[backlinks]]/`code` syntax) and the minimum word count are each independently configurable checkboxes in Settings, not fixed. Needs an API key configured above to do anything; if none is available right now — none set, or Secure Storage is locked so the saved one can't be read — the chip turns red and shows "paused," queued nodes are held rather than discarded, and the batch sends automatically the moment a key becomes usable again (e.g. right after unlocking) rather than waiting for the next edit to retrigger it. Clicking the chip while paused acts as the fix — unlocks Secure Storage directly if that's the cause, or jumps to the key field in Settings if not. That's separate from the retry-exhaustion case: if it fails three times in a row with a key that is available (e.g. an expired key), it turns itself off and tells you why, rather than silently retrying forever. Switching documents discards anything still queued rather than rewriting into the wrong document. Nodes created by Restructure text, Generate outline, paste-a-tree, and Word/OPML import are never queued for auto-rewrite regardless of this setting — only nodes you type into interactively go through commitEdit at all
Built-in providers — seven providers ship with Sakura: Gemini (free), Groq (free, fast), Claude API, ChatGPT (OpenAI), OpenRouter (free models), Cerebras (free tier), and GitHub Models (free, needs your GitHub account). This is the complete, fixed list — Sakura doesn't support adding another provider. All need an API key — selecting a provider shows a direct link to where you get one. GitHub Models requires a GitHub personal access token with models:read scope; model IDs in the dropdown already include the required vendor prefix (e.g. openai/gpt-4o)
Provider fallback — Settings → AI → Provider → Fallback order: enable the toggle to let Sakura automatically retry with the next checked provider when the active one fails — quota, rate limit, wrong model ID, or any other error. The active provider is always tried first; the list sets the retry order. Providers without a saved key are skipped automatically, and if none of the checked providers besides the active one have a key, a notice under the list says so directly rather than letting it fail silently the first time it's actually needed. A toast names both the failed provider and the one that succeeded (e.g. “Quota hit on Groq — rewritten with Gemini ✦” or “Error on Groq — rewritten with Gemini ✦”). The only error that does not trigger fallback is a bad API key (401) — that surfaces directly so you know to update it. Default order on a fresh install: Groq → Gemini → Cerebras → OpenRouter, with the first three enabled
To-Dos AI — Settings → AI → To-Dos AI, three independent toggles (each on by default): Extract action items adds a button in the To-Dos panel header that reads the current document — or your selected nodes, if any are selected — and proposes a batch of tasks from it, added directly with one Undo for the whole batch. Break into subtasks adds a small AI button on each open task (next to Rewrite) that proposes 3-6 concrete steps and adds them as real subtasks, also with Undo; it's hidden on tasks that already have subtasks or are set to repeat, since those can't hold subtasks. Status summary adds a header button that writes a short plain-language readout of your open tasks — what's overdue or high-priority first, then general progress — shown in a dialog with a Copy button. All three use the same provider/key/model as Rewrite above and need AI Capabilities on (Settings → Features); turning a toggle off here removes just that one button, leaving Rewrite and the other two untouched
Usage today — below the API key field, Sakura counts requests made to the currently selected provider today (and shows the count next to each provider in the Fallback order list once any requests are made). This is a local, best-effort counter — it resets on your computer's local calendar day, which can drift slightly from a provider's own reset clock (e.g. Gemini resets at midnight Pacific), and it isn't fetched from the provider itself. Treat it as a rough gauge of how much you've used a provider today, not an authoritative quota reading — check the provider's own dashboard (e.g. Google AI Studio) for the real number if it matters
Test key — the Test button next to Save sends a minimal request (asks the model to reply "OK") to confirm the currently typed or saved key actually works, without waiting for your next real Rewrite to find out. A toast reports success or the specific error. This check does not count toward the Usage today total above, since it's a connectivity check rather than real feature use

Utilities

Ctrl/Cmd + S → save now (kept in-app — doesn't trigger the browser's "Save Page As")
Print button in the header (next to Import ▾ / Export ▾) → prints the document with all UI chrome hidden; choose "Save as PDF" in the browser's print dialog to get a clean PDF. Checked nodes and tag chips render correctly in print; the file explorer, toolbar, and status bar are all hidden automatically
Export ▾ is grouped into Copy / Document / Data sections. Markdown, OPML, Text, and Print all respect Settings → Export & print → "Skip folded sections" the same way — folded branches are included or skipped consistently across all of them, not just whichever one you happen to use. PDF, PowerPoint, and Word are the exception: all three render from (or mirror) Preview/Presenter, which has its own independent fold state you control directly by clicking a section to expand or collapse it there — whatever that's currently set to is exactly what goes into the PDF, deck, or document, regardless of "Skip folded sections"
Import ▾ (next to Export ▾) is grouped into "From a file" (OPML, Word) and "From pasted text" (Restructure text, AI) — see the AI section above for how Restructure text and Word import handle structure
Ctrl/Cmd + Shift + D / Insert Date/Time in toolbar → timestamp
Copy Tree as Image → PNG
Ctrl/Cmd + Shift + C → copy selection
Clear → reset
Shortcuts that double as OS/browser bindings can occasionally be intercepted before they reach the page — if one doesn't fire, the matching button or menu item always works

Maximize / Zen mode

⤢ floating button bottom-right of the editor, Ctrl/Cmd + Shift + E to toggle it on or off, or Esc to exit only → by default hides the file explorer and app bar, so only the tree, toolbar (if it's on), tab bar, and a trimmed status bar remain
Settings → Zen mode → "Hide File Explorer" — on by default, so the file explorer auto-collapses when you enter zen mode. The header's own sidebar-toggle button (▤) keeps working normally throughout zen mode regardless of this setting, so you can always open or close it manually for that session — same "manual choice sticks until you exit" behavior as the toolbar and Pad below
Settings → Zen mode → "Hide status bar" controls whether the bottom status bar is hidden outright on entry — off by default. Left on, zen mode instead trims it down to just Words, Depth, Saved status, and the auto-rewrite indicator (plus Quick Assist, if it's normally docked in the app bar — see below), regardless of your normal status bar item selection, and restores your normal selection when you exit
Settings → Zen mode → "Hide tab bar" controls whether the row of open document tabs is hidden on entry — off by default, so you can switch between open documents without leaving zen mode. Turn it on if you'd rather it disappear too
A second floating button sits next to it — a shortcut for Settings → Bars & Menus → Toolbar → "Show toolbar," so it works the same way whether you're in zen mode or not. Click it, or Ctrl/Cmd + Shift + X, to show or hide the toolbar; the choice applies everywhere and is remembered next time
If you'd rather zen mode hide the toolbar by default, turn "Show toolbar" off before maximizing — everything it holds is still reachable by keyboard shortcut (see the other sections on this page) or by right-clicking a node for the icon row and "More" menu
The Pad panel has its own floating toggle next to the maximize button, since it stays open across zen mode
While actually presenting (not just browsing Preview), right-clicking a line's actual text opens a small popover built for capturing live feedback without breaking presenting flow — left-click always just advances to the next slide now, same as clicking anywhere else (matching the standard PowerPoint/Keynote/Google Slides convention with no exceptions); right-clicking anywhere that isn't a line's text still opens the laser-pointer toggle menu exactly as before. Four quick marks — ✓ Confirmed, ⚠ Issue, ⏸ Parked, 🚩 Follow-up — plus a fifth, ✗ N/A, for something discarded or out of scope, click one to set it, click the same one again to clear it (see the dedicated Markers entry above for what they do and how filtering works — the same marks are also available directly in the editor via right-click, without entering Presenter at all). Next to the marks, Edit turns that line's own text into a plain editable field right on the slide, in place — cursor already at the end, no popover involved at all; Enter or clicking away saves it straight back to the node, Escape reverts to what it said before. It's the same underlying text the editor and outline both use, so it's picked up everywhere the next time anything re-renders. Note… and Question… swap the same small popover's content in place for whichever one you picked instead — rather than opening a separate full-screen prompt; for something that's genuinely a line or two, a modal taking over the whole screen was more interruption than the content warranted. Each shows a compact header line (which one you're in, on the left; icon-only Cancel and Save on the right, plus a sparkle Rewrite (AI) button between them if AI Capabilities is on — Enter also saves, Escape also cancels) above a plain, borderless field, and both start blank and fold a short preview of which node you clicked into that same header line (label · node text), so you can confirm you've got the right one before typing without needing a whole extra row for it. Rewrite replaces the field's whole current text with an AI-rewritten version — there's no partial-selection rewrite here the way the main Pad editor has, since a plain text field doesn't carry that kind of selection state. Note appends a timestamped line to that node's Note field — if it already has note content, a Delete icon appears next to Save so you can clear the whole note in one action instead of only ever adding to it. Question adds a new Q&A row linked to that node — neither Note nor Question copies the node's own text into what gets saved, only what you actually type; the Q&A row links back via the same mechanism a Scan-created row uses for its own link icon, so nothing about the outline itself gets duplicated. Either way, a small indicator dot appears on that line (both in Preview/Presenter and back in the editor's own outline) — the same note-dot an ordinary note gets, or a dedicated question-mark dot for a Question — so a captured line stays visibly marked rather than only living inside the Pad; click either dot to jump straight to what was captured
Settings → Zen mode → "Hide app bar" also hides the very top row (Documents/Templates, search, Settings, Help, Import/Export) while zen — on by default. If Quick Assist is normally docked in the app bar ("Dock in app bar" in Settings → Quick Assist), zen mode relocates it into the status bar for as long as zen mode is on instead, rather than popping the app bar back open every time you invoke it; your "Dock in app bar" preference itself is untouched and Quick Assist returns to the app bar when you exit zen mode. If the app bar is hidden, right-click the editor canvas for a "Show app bar" option (only appears while it's actually hidden), or use the keyboard shortcuts for whatever you need — everything the app bar holds still works that way too
Settings → Zen mode → "Use browser fullscreen" — on by default — additionally requests the browser's own fullscreen (the same as pressing F11) when you enter zen mode, hiding the browser's tabs and address bar too, not just Sakura's own UI. Exiting fullscreen — Esc, F11, or the browser's own control — exits zen mode at the same time, so the two never end up out of sync

Preview

floating button bottom-right of the editor (next to zen/Pad/toolbar) → opens a read-only, presentation-style render of the current document. Nothing here can be edited, and closing it never changes your actual document
Headings come from the Heading 1–6 style you apply in the editor (Format toolbar, next to Text color) — not from tree depth. A node with no heading style of its own renders as plain body text in Preview, unless Settings → Preview → "Headings & Contents" has a level assigned to that node's depth (see below)
Settings → Preview → "Headings & Contents" → "Heading fallback by depth" lets you assign a heading level to depths 0–5, so an unstyled node can still render as a heading in Preview based purely on how deep it sits in the tree. An explicit heading style on the node always wins over this; depths left as "None" behave exactly as before. This only changes how Preview renders — it never writes a heading style into the actual document
Table of contents (left panel, toggled with the header button) is built from headings and [Section] markers — what it includes is configurable in Settings → Preview → "Headings & Contents": cap it to a maximum heading level, or turn off section markers to keep it to real headings only. "Treat depth 0 as section markers" there (off by default) makes every top-level node act as a [Section] divider automatically, without needing the [Bracket] text — [Bracket] text still works at any depth on top of this, including depth 0. Unlike the Table of Contents toggle, this one changes the document itself: the outline canvas, Preview/PDF, and rich-text copy all pick up the same section styling, not just what's listed in Contents. Below the headings, a "Decisions" section lists every decision log in the document with a colored status dot — click to jump straight to it. The panel shows itself if there are 2+ headings or at least 1 decision log, whichever applies; it's part of what gets printed to PDF, so the decision index carries over there too
Presenter mode (▶ button in the header) goes fullscreen and hides all the header buttons — the bottom bar (prev/next, slide counter, exit) is the only control surface while presenting. Opens on a decorated title slide (document title, author, the same word-count/decision-count line as the Preview header) before stepping through the rest one section at a time — arrow keys/space to navigate, clicking the slide itself also advances (same as PowerPoint/Keynote), Home/End to jump to first/last, Esc (or leaving fullscreen any other way) to exit. If you've zoomed in with Ctrl+scroll, Esc resets the zoom first instead of exiting — but the browser itself still treats Esc as "leave fullscreen" regardless (a security behavior no page can override), so that first Esc press also drops out of true fullscreen even though it doesn't leave Presenter mode; presenting just continues windowed after that until you exit and re-enter Presenter mode (an earlier attempt at an 'F' shortcut to restore fullscreen manually didn't reliably work and was removed rather than left half-working). Fold state, notes, code blocks, and decision cards all render normally within each slide. Each slide fades in briefly on navigation (not on every edit — only an actual slide change), a thin progress line tracks your position across the top of the screen alongside the slide-counter text, and headings/sub-bullets get a size and weight bump specific to presenting (sub-bullets step down in size by depth on top of the existing color-dimming, headings get bolder) so the hierarchy reads faster from across a room — none of this changes how the same content looks in the regular scrolling Preview. A running elapsed timer sits in the bar. G opens a grid of every slide's label — click one to jump straight to it, useful for Q&A on longer decks; Esc closes the grid without exiting presenter mode. Right-clicking anywhere that isn't a line's own text opens a small menu: Reset zoom and Laser pointer are always there (same as in the regular scrolling Preview), and while actually presenting it also gets All slides, Notes / Q&A, Whiteboard, Mind Map, and Exit presenter mode — the same actions as their keyboard shortcuts and buttons, just discoverable by right-clicking if you don't remember (or haven't learned) those shortcuts. Each of those five hides itself automatically if its underlying feature is off in Settings, same as the presenter-bar button it mirrors. A two-monitor icon in the presenter bar (or S, and Open/Close Audience View in the right-click menu) opens a second, independent browser window showing this same document — normal editor chrome (file explorer, toolbar, everything) stays hidden there, leaving just the current slide — meant to be dragged onto a projector or a second monitor while this window stays as your own presenter view. Since Presenter mode fullscreens this window by default, opening the Audience View drops this one back out of fullscreen first — the point is for the *audience* window to be the one that's fullscreen on the projector, not both at once — presenting itself isn't interrupted, this window just continues windowed afterward the same way an Escape-triggered fullscreen exit already leaves it. The new window shows the slide right away with a small banner at the bottom — drag the window into place, then click that banner (or press F) to make the window fullscreen there — browsers require that click to happen in the new window itself before it's allowed to go fullscreen, so this can't be skipped automatically. Beyond that one click, it's a pure display — clicking the slide there or pressing arrow keys in that window does nothing, so it can never drift out of sync with what you're actually presenting. From then on it stays in sync automatically — slide changes, B blank-screen, the laser pointer, any Ctrl+scroll zoom adjustment, and even a live content edit made while presenting all mirror across within the same navigation that already updates this window, with no separate export or file involved, since it's just a second view of the same open document. With the Audience View connected, a "Next: …" label also appears next to the slide counter in this window's own bar, so you can see what's coming before you advance to it — hidden the rest of the time, since there's nothing to look ahead of if there's only the one screen. Opening the Whiteboard or Mind Map (below) while the Audience View is connected swaps its display over to that too, the same way it takes over this window's own screen — both mirror live: Mind Map because it's plain on-page content same as slides, Whiteboard by pushing each save straight into the Audience window's own draw.io session (see the Whiteboard entry below). Markers, per-node notes, and Q&A questions added to a node while presenting also carry across, on the same refresh that already updates the slide when its content changes. Closing that window, pressing S again, or exiting Presenter mode (which also closes it automatically) all disconnect it the same way; if it's ever popup-blocked, a toast says so rather than silently doing nothing. B blacks out the screen to pull attention back to you mid-discussion — press again (or Esc) to resume exactly where you were, nothing about the slide itself changes. N opens a floating Notepad panel without leaving Presenter mode — Q opens that same panel straight into its Q&A view instead (pressing the key for whichever view is already showing closes the panel; the other key switches views without closing it) — it's the same Pad that document already has (Settings → Panels → Pad), the real element (editor and toolbar both) just moves into view and back, so whatever you jot down while presenting is that document's Pad afterward, nothing separate to lose track of; needs Pad enabled and a document open. Once it's open, its header also has two small Notes / Q&A tabs you can click directly — the same switch as the N/Q keys, just visible for anyone who'd rather click than remember a shortcut. The panel is movable and resizable — drag its header to reposition it anywhere on screen, drag the small grip in its bottom-right corner to resize it, double-click either one to snap back to the default docked position — whatever you leave it at is remembered next time you open it, this session and future ones. A small eye icon in its header shows this same Notepad/Q&A view on the Audience screen too, if it's connected — off by default, since notes are usually your own prompts rather than something meant for the room; toggling it swaps the Audience window's display over to it the same way Whiteboard/Mind Map do, and switching between the Notes/Q&A tabs while it's on updates the Audience window to match. It turns itself back off if you close the panel, exit Presenter mode, or disconnect the Audience View. Separately, per-node notes (the small dot that appears on a node with a note attached) stay collapsed on their slide until clicked — clicking one opens it in a popover without leaving Presenter mode. That's a different feature from the floating Notepad panel above it — despite both involving the word "note," one is the whole-document Pad floating over your slides, the other is notes attached to specific nodes and rendered inline within them. A pencil icon in the presenter bar (or W, and Whiteboard in the right-click menu) opens one persistent draw.io diagram for the document — the same real, embedded draw.io editor the Diagrams tab uses elsewhere, not a separate drawing tool — for sketching or brainstorming mid-talk without it needing to be tied to any particular slide or node; the first click creates it, every click after reopens the same one, and it shows up pinned at the top of the regular Diagrams tab list too, marked with a "Whiteboard" badge instead of the usual "Unlinked" warning, since staying unlinked is the point rather than something to fix. Closing it while presenting (Esc, or the close button) is instant, same as Notes/Mind Map — the "saved this already?" confirmation that normally guards against losing unsaved draw.io changes only appears outside Presenter mode, since a confirmation dialog interrupting a live talk is worse than the rare case of losing an unsaved doodle on the scratch whiteboard. If the Audience View (above) is connected, opening the Whiteboard opens the same drawing in its own live draw.io session there too (loaded fresh from this document's saved copy the moment the Audience View connects), and every save here pushes the latest drawing straight into that session — genuinely live, not a periodic snapshot, since both are real draw.io editors and this uses the same load mechanism draw.io's own embed protocol already provides. Needs the Diagrams feature enabled (Settings → Features → Pad Panel) — the button and menu item both hide themselves if it's off, same as everywhere else in Sakura that depends on a feature switch. A second icon next to it (or M, and Mind Map in the right-click menu) opens the app-level Mind Map canvas without leaving Presenter mode — the same canvas its own status bar button and shortcut open elsewhere, not tied to this document or slide; hides itself if the Mind Map feature is off in Settings. The floating panel shows its own, smaller set of toolbar buttons — Bullet list and AI Rewrite by default — independent of the full toolbar the regular Pad panel shows; pick a different set in Settings → Preview → Presenter mode → Toolbar buttons while presenting. Timer, the all-slides grid, and notes can each be turned off in Settings → Preview → Presenter mode, along with which tree depth starts a new slide, auto-enabling the laser pointer, "Open Preview straight into Presenter mode" (skips the regular scrolling Preview — opens straight to the title slide, fullscreen, and Esc then takes you all the way back to the editor in one press instead of backing out one layer at a time), a small optional branding mark in the presenter bar (and, from the same setting, in the corner of exported PowerPoint slides) — "SAKURA" by default, or your own company/team name via "Branding text" — off by default, and a closing slide (on by default) — one more slide after the last content slide, an accent bar plus a short line styled to bookend the opening title slide, reading "Thank you" unless you set your own text and optional smaller subtitle underneath in the same settings section
Settings → Preview → "Reading appearance" adds reading width, a sepia reading theme, and a serif body-text option — Preview-only, doesn't touch the editor or document. Settings → Data & Backup → "Export & print" → "PDF page margins" sets the default Export as PDF starts with; the print dialog that opens still lets you change it per export
Click ▾/▸ → fold/unfold a branch, same as the editor — but this fold state is local to Preview and separate from the editor's own collapsed sections. Collapse all / Expand all buttons sit in the header
Folding a branch that contains a note or a decision log doesn't hide it silently — a small icon next to the +N badge tells you it's in there; click either to expand
Notes render as their real rich content (images, lists, links) inside a card, not the small dot from the editor
Decision logs become a card with a "Decision Log" header — Status always shows there as a colored badge — including its unedited default of "Proposed", since that's a real status rather than a placeholder question (color is guessed from the word itself: approved/accepted/implemented/resolved/done/confirmed → green, rejected/declined/blocked/abandoned/cancelled → red, hold/pending/review → orange, anything else → gray). Each other answered field gets its own labeled block below, in order — and if you've written the actual answer as child nodes under a field (e.g. a numbered list under Rationale, or bullet points under Impact) rather than on the field's own line, that nested content renders as a list inside the block, numbered automatically if your items start with "1.", "2.", etc. A field only shows if it has real content — either its own line has been edited past the default question, or it has children. Which fields appear at all is configurable in Settings → Preview → "Decision card fields" (Context + Decision by default; this only affects how Preview summarizes a log, not what the Decision Log feature itself inserts)
Right-click anywhere in Preview → Laser pointer, for pointing at things while presenting without your real cursor in the way. Turns off automatically when you close Preview
Export buttons in the header (PDF / Word) are the same export as everywhere else in Sakura — Preview just gives you a one-click path to them while you're already looking at the rendered result

Large trees

Click ▾/▸ → fold / unfold subtree
← / → → collapse / expand selected
Ctrl/Cmd + F → search this tree; Ctrl/Cmd + H → search and replace (Replace / Replace all, undoable)
Ctrl/Cmd + Alt + C / ▶ in toolbar → collapse all (folds every top-level section)
Same shortcut / same button again → expand all — it's a single toggle, the icon flips to show which one it'll do next

Documents & folders

button in file explorer header → new folder; button → new document (unfiled); or Ctrl/Cmd + Alt + N from anywhere
Hover a folder → ＋ creates a document already filed there; ✎ renames it; deletes it (subfolders move up a level, documents move to Unfiled — Undo appears briefly after deleting)
Right-click a folder → new document or new subfolder here, rename, or delete
Right-click a document → rename, duplicate, move to a folder, or move to Trash
Click ✎ / ⧉ / icons on a document row → rename / duplicate / move to Trash; the folder-icon button moves it via a dropdown list
Click a document's icon (left of its name) → pick from a small set of icons to mark it visually, or "Use default icon" to clear it back to the plain page icon. Purely visual, doesn't affect the document's content
Click a document's or folder's name → switch to it / expand it
Folders can be nested: drag a folder onto another folder's top or bottom edge to reorder it as a sibling there, or onto the middle to nest it inside
Drag a document onto a folder (or "Unfiled") to file it there; drag it onto another document to reorder within that folder/Unfiled
/ buttons by "Documents" → collapse all / expand all folders (only shown once a folder exists)
Ctrl/Cmd-click a document to multi-select; Shift-click selects a range. A bar appears to move or trash the whole selection at once
Deleting a document (or a selection of them) — or a template, including a built-in one — moves it to Trash, a collapsible section at the very bottom of the file explorer, below Templates — out of the way since it's rarely needed, same as most apps put it. It's not gone. Each item there shows when it was trashed, with buttons to restore it or delete it forever; documents and templates sit in the same list, told apart by their icon and a small "Template" label. A toast with "Undo" also appears right after trashing, for a quick reversal without opening Trash at all. Settings → Trash controls how long items sit there before being purged automatically (default 7 days; "Forever" turns off auto-purge so only "Empty Trash" removes anything) — one setting, shared by both. Header search (Ctrl/Cmd + K) skips Trash by default — turn on Settings → Trash → "Include Trash in search" to have it surface matches there too (across documents, templates, node text, Notes, tags, and the Pad alike), clearly marked "In Trash"; clicking one reveals it in the Trash section rather than opening it directly, since restoring is a deliberate choice. Settings → Data → "Restore built-in templates" also pulls a trashed built-in back out of Trash, not just refreshes its content
Searching the file explorer keeps folders visible (showing only matching documents and subfolders, force-expanded) but disables drag-and-drop while a search is active; it only matches document/template names — for text inside documents, use the header search (Ctrl/Cmd + K) instead
The file explorer's filter box is hidden by default since the header search now covers the same ground; bring it back via the icon next to "Documents", or Settings → Bars & Menus → File Explorer → Filter box
Drag the thin strip at the file explorer's right edge to resize it
Version history (right-click a document → "Version history", or "More" → "Version history" for the one currently open) keeps automatic checkpoints of a document's past content — not full version control, just lightweight snapshots. A version is captured right before it would otherwise be overwritten, but only once every ~10 minutes of active editing at most, and only if something actually changed, so it won't flood with near-duplicates from every autosave tick. Capped at the last 20 per document, oldest dropped first. "Save a version now" in that panel adds one immediately, useful right before a big rewrite. Restoring an older version is itself non-destructive — whatever was there gets saved as a version first, so restoring is never a one-way trip
If the same document is open as the active one in two tabs or windows of this browser at once (easy to do by accident — duplicating a tab, browser session restore, opening the file twice), a banner appears the moment the other one saves a change, since otherwise whichever tab saves last would silently win with no warning. "Reload latest" discards anything unsaved here and pulls in the other tab's version; dismissing it just clears the banner — your edits here are kept either way, and the other tab gets the same warning in reverse once you save. Documents open here only in the background (not the one on screen) refresh silently instead, since there's nothing unsaved there to protect. This only applies within one browser — a completely different browser, or a different device, never shares storage and can't collide

Templates

Click a template's name in the Templates section → load it into the current document (replaces current content; Undo restores it)
Default → loads your custom default template, if one is set
⋯ button → Save current as template, or Set current as default
button next to ⋯ → new template folder, for grouping templates (single level; templates can be moved between folders via the move-to-folder button or right-click, but folders themselves can't be nested or reordered)
/ buttons by "Templates" → collapse all / expand all template folders (only shown once a template folder exists)
Click a template's icon → same small icon-picker as documents, to mark it visually. Built-in templates keep their fixed icon (grouped by document shape) and aren't customizable this way
Ships with a built-in pack — Decision Record, plus a set of SAP architecture templates — grouped in a "Built-in" folder. (Meeting-style templates like Meeting Notes, 1:1s, Daily Standup, Weekly Status, Project Kickoff, and Retrospective live inside the Meetings panel's own "New from template" menu instead, since they fill in a meeting note rather than insert outline nodes.) The folder and the templates in it can be renamed and moved like any other, but are deleted like any other — deletion goes to Trash and is reversible. Use the ↺ restore button in the Templates file explorer header to restore any deleted or renamed built-in templates at any time. Icons are grouped by document shape, not unique per template
The "Templates ▾" button in the header toolbar opens the same template list as a quick-access dropdown, grouped by folder exactly like the file explorer (icons included) — useful when the file explorer itself is hidden, e.g. in zen mode. It's read-mostly there: click a template to load it, plus rename/move-to-folder/delete per row, but folders themselves are only created from the file explorer
A line that's exactly "Date:", "Author:", or "Date · Author" in any template (built-in or your own) gets auto-filled the moment the template loads — today's date, and the document's current author if one is set (see Settings → Editing → "Default author" to pre-fill that automatically too). The merged "Date · Author" form fills in with no labels, just the values
Built-in templates use one color convention throughout, worth keeping if you duplicate and extend one: purple = neutral section header, orange = action/decision, red = risk/blocker, green = positive
Want a template in a new doc instead of overwriting? Create a new document first, then load the template

Tabs

Click a document in the file explorer, or the + at the end of the tab strip → opens it as a tab. Switching tabs keeps each one's undo history, scroll position, and selection independent — it doesn't reset like a plain document switch would
Double-click a tab → rename it
on a tab, or Ctrl/Cmd + Alt + W → closes the active tab only; the document itself isn't deleted and stays in the file explorer. Closing the last tab leaves the editor empty rather than reopening another document
Drag a tab left or right to reorder the strip
▾ at the right of the tab strip → search all open tabs by name; ↑ / ↓ + Enter to jump, or click a result
"Reopen tabs on launch" (a Quick Assist toggle — search "reopen tabs") controls whether your open tabs come back next time, or you start with just one

Backing up your data

Everything is stored locally in the browser, scoped to this exact file. Opening a different copy or a newer version of this file starts with empty storage — that's a browser limitation, not a bug
Settings → Data → Auto-backup to file → keeps a live backup written to a file on disk as you work (Chrome/Edge only — Safari and Firefox don't support this yet). Browsers revoke write permission to that file across a reload or restart as a deliberate security measure, so this can need reconnecting — a chip in the status bar flags it when that happens, and clicking it (or "Reconnect" in Settings) restores it in one click. If you disconnect it yourself from Settings, that same chip stays visible with a plain "Connect backup" prompt rather than disappearing, so there's always a way back to it from the status bar. Since the live file always overwrites itself, a "Backup history" list appears underneath once it's written at least once — up to 5 timestamped snapshots, at least 15 minutes apart so a burst of edits doesn't just fill it with near-duplicates, each independently restorable if a bad edit gets written to the live file before you notice
Settings → Data → Local safety copy → a second copy kept automatically in a separate browser storage area; Restore brings it back if the main storage gets cleared. Not a substitute for Export — it's still inside the same browser
Settings → Data → Undo last restore → appears only after you've used Restore (from a backup file, the Local safety copy, or a Backup history entry above) — a snapshot of whatever was in the app right before that restore, in case the file you picked turns out to have been the wrong one. One level deep: it holds only what was there before your most recent restore, not a full history of every restore you've ever done
Settings → Data → Secure Storage → optional, encrypts AI provider keys and the GitHub Gist token at rest (AES-GCM, passphrase-derived key) instead of storing them as plain text — in this browser and in local backup files. The passphrase itself is never stored anywhere; unlocking derives the key fresh each time, and it only lives in memory for that session — reload the page and you'll need to unlock again to use AI features or Gist auto-sync. There's no recovery if you forget it. Turning it off writes everything back as plain text. Once it's set up, a status-bar chip shows the current Locked/Unlocked state and toggles it in one click — no need to open Settings each session; right-click the chip to jump to the full Secure Storage controls instead. Hide it like any other status-bar item in Settings → Statusbar items if you don't want it there. Unlocking also automatically resumes anything that was paused waiting on it — Auto-rewrite on commit and Gist auto-sync each pick back up on their own the moment you unlock, with no need to redo whatever triggered them originally
Settings → Data → Cloud Backup → optional, pushes the same full backup as "Backup all data" to GitHub Gist or Google Drive instead of a local file, so you can pull it down again from a different browser or computer. GitHub Gist needs a Personal Access Token with the gist scope, pasted into Settings — stored in this browser only, the same as an AI provider key (and covered by Secure Storage above, if that's turned on). Google Drive needs your own free OAuth Client ID from Google Cloud Console; the app only ever sees files it created itself, not the rest of your Drive, and the sign-in token lives in memory only — you reconnect each session, nothing Google-related is written to disk. Both providers find your existing backup automatically by name/description on Save and Load, so there's no ID to copy between browsers by hand. Both are manual push/pull by default — Save fully overwrites the cloud copy with what's here, Load fully overwrites what's here with the cloud copy, no merging either direction. Both also offer "Auto-sync on edit" (off by default) — pushes automatically ~1.2s after edits settle, same timing as Auto-backup to file. Gist's auto-sync persists across reloads (the token is saved) and additionally detects conflicts: if the Gist changed elsewhere since your last Save/Load here, auto-push pauses itself and shows a conflict notice rather than silently overwriting the newer version — resolve it with a manual Save or Load. Drive's auto-sync is session-only, since Drive's sign-in token isn't kept across reloads — it turns itself off with a one-time notice (not a nag) whenever the session connection is gone, and needs reconnecting and re-enabling next time; it has no conflict detection of its own, so treat it as a convenience rather than a safety net if you also edit from Drive on another device. Gist's auto-sync also pauses (with a one-time toast, not a nag on every edit) if the token itself can't be read because Secure Storage is locked — nothing is lost in the meantime since local saves and the safety copy never depend on the vault, and the paused push resumes and sends on its own the moment you unlock, no extra edit needed to retrigger it. Neither AI provider keys nor the Gist token — encrypted or not — are ever included in what gets uploaded to Gist or Drive; only your documents are
Settings → Data → Outlook Calendar → optional, lets the Meeting Notes panel's "Import from Outlook" button (calendar-with-check icon, top toolbar) pull events straight from your Outlook calendar instead of typing meeting details in by hand. Needs your own free app registration from Azure Portal (a Single-page application, with this app's URL added as a redirect URI and Calendars.Read permission granted) — same "bring your own OAuth client" shape as Google Drive above, and the same reasoning: no backend server, sign-in token lives in memory only for this session. Read-only — importing never writes or changes anything on your actual calendar. Import shows a checklist of events from a few days back through about two weeks ahead (recurring events expanded into their real occurrences, not just the series); each creates a new meeting note pre-filled with title, date, time, attendees, and location. Already-imported events are shown greyed out and pre-unchecked (matched by the event's own ID) so re-opening the importer later won't create duplicates of the same meeting. The document icon next to it, Import from .ics file, needs none of that setup — pick any .ics file (exported or shared from Outlook, Google Calendar, Apple Calendar, or elsewhere) and it's parsed entirely in the browser, no sign-in or network call at all. Same checklist, same pre-filled fields, same already-imported dedup, matched by the file's own event UID instead of a Graph API event ID. One real gap: an event whose recurrence is more than a simple daily or weekly pattern (monthly, yearly, or anything with a by-month-day/by-position rule) imports as just its one base occurrence, not expanded into every future instance the way the Outlook API import does — the picker flags any such event so that's never a silent surprise. Plain daily/weekly recurrences (including "every N days/weeks" and "every Mon/Wed/Fri") are expanded into their real upcoming occurrences, up to about six months out
Settings → Data → Backup / Restore → still the way to move data between files, browsers, or computers, or to take a one-off manual backup; the file explorer footer has Backup/Restore shortcuts too, so you don't need to open Settings just for that. The header's Export ▾ and Import ▾ buttons are for the current document only (Markdown, Word, OPML, etc.) — Backup/Restore cover all your data
Settings → Data → "Warn if this session isn't backed up" → offers Export or Connect auto-backup if you've made changes this session but neither has run yet. Three modes: Strict shows the browser's own "leave site?" prompt when you try to close the tab, then this dialog if you cancel it — a real last-chance gate, at the cost of that generic browser popup (its wording is fixed by the browser, not by Sakura). Passive skips the browser prompt entirely and shows only this dialog, the next time you switch away from or close the tab — quieter, but a direct close with no prior tab-switch gets no warning at all. Off disables both; the status-bar "⚠ Back up your data" chip (from the reminder setting above) is your only remaining signal
Recommended: export before switching to a new version of this file, then import right after opening it
Settings → Data → Reset to default settings → restores appearance/behavior preferences only (theme, toolbar, right-click menu, etc.); your documents and AI settings aren't touched

Note panel

Every node can hold a rich-text note and, on its own tab, a code block — open the panel via the toolbar ( Note button), right-click → More → Note, or Ctrl/Cmd + Shift + N (jumps straight to the Note tab; Ctrl/Cmd + Shift + K does the same for Code) — clicking the chat-bubble dot that appears on a node with either also opens it directly, defaulting to whichever tab has content (Note first if both do)
Note and Code share one floating, draggable window with a tab switch in its header — one dot on the node row instead of two, and no separate toolbar button or right-click entry for Code. A small </> badge on the dot signals the node also has code, even while you're looking at the Note tab. Drag the header to reposition the panel anywhere; click to expand to full-screen and click again to restore — maximize applies to whichever tab is active
Note and Code open in a minimal Normal view by default — just the editor, with the toolbar, backlinks, and timestamps tucked away and Rewrite relocated up into the header, since most notes are a short remark rather than something needing the full toolbar. Click to expand to Maximize — full toolbar, footer, and backlinks/timestamps all visible, Rewrite back in its usual toolbar spot — and click again to restore. Maximize applies to whichever tab is active
Two small badges in the header — a clock and, only when the node is actually referenced elsewhere, a link icon with a count — reveal timestamps or backlinks inline in Normal view without leaving it, so they're reachable without a full Maximize. Both start collapsed each time the panel opens and hide automatically in Maximize, since everything's already showing there. Neither badge appears on the Code tab, which has no equivalent
Toolbar: B I U S for bold / italic / underline / strikethrough; bullet list and numbered list; insert or edit a link, with its own display-text field separate from the URL; table icon inserts a 2×2 table; AI Rewrite rewrites the whole note or just selected text; the summarise button reads the node's child subtree and appends a prose summary to the note
Tab → indents a list item (when inside a list) or inserts spaces in plain text; Shift + Tab → outdent a list item
Highlight any text in the note (Normal or Maximize view) and a small formatting popover appears above it — Bold, Italic, Strikethrough, Highlight (a yellow inline highlight, click again on already-highlighted text to remove it), Link, and Rewrite (AI, rewrites just the highlighted phrase — same underlying call as the toolbar's own AI Rewrite, just pre-scoped to the selection). A quicker route to the same commands as the toolbar above, without moving the cursor up to reach it. Settings → Editing → "Selection formatting popover" turns this off
Rewriting with nothing selected replaces the note's entire content with plain AI-rewritten text, which would otherwise silently remove any images or tables in it — a confirmation prompt catches this and tells you what would be lost before it happens. Select just the text you want rewritten instead if you'd rather keep the rest untouched
Paste images directly into the note — they embed inline and are stored with the note. Click a pasted image to select it, then drag the handle at its corner to resize; double-click it to reset back to its original size
Feature as diagram — right-click any image in a note and choose "Feature as diagram" (e.g. for architecture diagrams) to have it pulled out of the small collapsible note card in Preview and Presenter mode and shown instead as a large, always-visible figure right under the node's text — no click needed to reveal it. The image's alt text (if it has one) shows underneath as a caption. Right-click it again to turn this off and return it to normal note behavior. This also applies to PDF exports, since those render from the same Preview output
Tables: drag the thin border handle between columns to resize column widths
Ctrl/Cmd + Z / Ctrl/Cmd + Y undo and redo edits made in this note while it's open — its own undo history, separate from the outline's and separate from the Code tab's, reset each time you open a note and cleared when you close it. The whole session still counts as one step on the outline's own undo when you close the panel. Ctrl/Cmd + Enter or Esc → close and save (whichever tab is active); "Clear note" removes the note text (undoable) and closes the panel
A chat-bubble dot on a node row signals it has a note and/or code — click it to open the panel on its default tab; this also selects that node
Backlinks — if other nodes reference the current node via @mention, a Backlinks section appears at the bottom of the Note tab. Click any entry to jump to that referencing node
Timestamps — the Note tab also shows when the node was created and last modified, on a single line at the bottom. Existing nodes show the time they were first loaded after this feature was added; new nodes and any node whose text is edited going forward will have accurate timestamps

Code (in the Note panel)

Every node can also hold one code block, on its own Code tab inside the same panel as its Note — there's no separate popup, toolbar button, or right-click entry for it. Switch tabs with the header switch, or jump straight there with Ctrl/Cmd + Shift + K. A node with code but no note text opens directly to the Code tab when you click its dot
A real <textarea> for editing, syntax-highlighted live as you type — no rich-text formatting, just plain code with color. Line numbers down the left edge. Drag the bottom edge to resize it vertically. Dragging the panel, maximizing it, and closing it all work the same as described above for the Note tab, since it's the same window
Language picker: Plain text, ABAP, SQL, JavaScript, Python, JSON, XML/HTML, Markdown
Copy copies the raw code text (not HTML) to your clipboard
Ctrl/Cmd + Z / Ctrl/Cmd + Y undo and redo edits made in this code block while it's open — its own undo history, separate from the Note tab's, reset each time you open a code block and cleared when you close it. The whole session still counts as one step on the outline's own undo when you close the panel. Ctrl/Cmd + Enter or Esc → close and save; "Clear code" removes it entirely (undoable) and closes the panel
A small </> badge on the node's note dot signals it has a code block, whichever tab you last had open

Pad Panel

A docked panel for whole-document content, separate from per-node Notes — one Pad per document, not tied to any single node. Six tabs live inside it: Notepad, Q&A, Decision Log, Diagrams, Mind Map, and Files — each documented in its own card below
Open it with the small panel icon docked at the bottom-right of the editor, next to the maximize button, or Ctrl/Cmd + Shift + P; the same shortcut and button close it again. It stays open if the editor is maximized, and that floating button is how you open or close it while maximized too — handy if Settings → Bars & Menus → Toolbar → "Show toolbar" is off and the regular toolbar isn't around to use instead
Drag the thin strip on its left edge to resize it; width is remembered, as is whether it's open or closed
The header bar (the seven tabs above, plus the close button) is always there — it's how you switch modes. "Match editor background" (Settings → Pad) controls whether Pad's writing surface looks like the outline's canvas or reads as visually distinct chrome — personal preference, no right answer. Settings → Pad → "Default pad view" picks which tab greets you every time you open the Pad — you can still switch between them with these same header tabs while it's open
Settings → Features → Pad Panel lets you turn off individual tabs you don't use (Q&A, Diagrams, Mind Map — Notepad and Decision Log have their own toggles alongside it) as well as the whole panel at once

Notepad

The Pad's default tab — a rich-text scratchpad for the whole document, using the same toolbar as per-node Notes
Toolbar: B I U S, bullet / numbered lists, link (with its own display-text field, separate from the URL), table insert with resizable columns, image paste/insert (click to select, drag the corner handle to resize, double-click to reset), and AI Rewrite (rewrites the selection, or the whole Notepad if nothing's selected). No subtree-summarise here — there's no subtree at the document level. Settings → Pad → "Toolbar buttons" hides individual ones you don't use
Tables — the toolbar button inserts a 2×2 table with a real header row; Tab inside any cell moves to the next one (spreadsheet-style), Shift+Tab moves back, and tabbing forward from the very last cell appends a new row and lands in it. Right-click a cell for the full table menu: Add row below, Add column right, Delete row/column/table, and Toggle header row — flips the first row between a plain row and a real header (bold, repeats as a header in Word/PDF/PowerPoint), useful for a table pasted in from elsewhere that didn't arrive with one. Below that, a Presentation group: Feature under selected node links this table to whichever node is currently selected in the outline, so it appears (as a genuine, editable table) in that node's Preview, Presenter, PDF, Word, and PowerPoint output — an accent outline marks a featured table so it's clear at a glance, and re-running it against a different selected node re-links rather than duplicates. Once featured, Feature as chart becomes available (shown disabled with the reason on hover if the table's shape isn't chartable — 6+ columns, or no numeric column to plot) and swaps that node's output — in Preview, Presenter mode, PDF, and PowerPoint alike — from the table itself to a real chart built from it: column A is read as category labels and every column after it as its own data series (so 3-5 columns becomes a grouped/multi-line chart), except a single-column table, where the one column's own numeric values become the series and each row is labeled by its position. Toggle it again to go back to exporting as a plain table. Once charted, more items appear for picking the chart's actual type — Bar, Line, or Pie (pie uses the first data column only, since a pie can't meaningfully represent more than one series), or Cards, which lays the table out as a row of stat cards — one per row, each showing its value alongside its label — instead of a plotted chart; like pie, it only applies to a single data column, and each value shows exactly as typed rather than being parsed as a number, so a non-numeric value like "<2s" displays correctly. Cards is a Preview/Presenter/PDF-only display — PowerPoint export falls back to a real table for a Cards-featured table, since there's no native PowerPoint chart type for it. Once showing as Cards, a Suggest icons with AI item appears, which sends every card's label and value to the AI in a single request and picks one icon per card from a small fixed set (security, speed, time, scope, cost, people, growth, quality) based on what each card is actually about, rather than the plain keyword match Cards falls back to automatically when nothing's been suggested yet. The current type is shown disabled rather than with a checkmark, same "already set" treatment used elsewhere in this menu. The same three options are also reachable by right-clicking the rendered chart itself in Preview or Presenter mode, without coming back to the Notepad first — that same right-click menu also has a Show as table item at the bottom, the same on/off flip as unchecking Feature as chart back in the Notepad
Running AI Rewrite with nothing selected replaces all of Notepad's content with plain rewritten text — which would silently remove any images or tables it contains. A confirmation prompt now catches this case and tells you what would be lost before it happens; select just the text you want rewritten instead if you'd rather keep everything else untouched
Highlight any text and a small formatting popover appears above it — Bold, Italic, Strikethrough, Highlight (a yellow inline highlight, click again on already-highlighted text to remove it), Link, and Rewrite (AI, rewrites just the highlighted phrase — same underlying call as the toolbar's own AI Rewrite, just pre-scoped to the selection, with the same confirmation logic not applying since a selection is already guaranteed here). Settings → Editing → "Selection formatting popover" turns this off
The toolbar's Quote button inserts a styled block for quoting or citing something — select text first and it wraps that selection (cursor lands on the citation line right after, ready to type a source); with nothing selected it drops in a "Quoted text" placeholder that's pre-selected so typing replaces it immediately. Renders italic with a left border
Has its own undo history, separate from the outline's — Ctrl/Cmd + Z / Ctrl/Cmd + Y undo and redo typing, AI Rewrite, and "Clear pad", for the current document. It never appears on the outline's own Ctrl/Cmd+Z stack, since this content isn't part of that snapshot — the two undo histories are independent, and this one is per-document, cleared on reload the same way the outline's is. For anything further back than that session's undo history, Version History (More menu → Version History) also keeps periodic snapshots of Notepad text — restoring one brings back the text (not pasted images, which aren't kept in History)
Settings → Pad → "Notepad" toggle hides this tab and its floating dock toggle entirely if you don't use it (on by default)

Files

One of the Pad's tabs — a plain list of files attached to the whole document, separate from the Notepad's own text (an earlier version of this feature inserted attachments as inline chips inside the Notepad itself; they now live here instead, as their own list)
Click the + toolbar button and pick any file, or drag a file onto this tab — either adds a row (with a file-type icon — PDF, image, spreadsheet, doc, slides, archive, audio, video, or code, based on extension) showing the filename and size. Click a row's name to download that file again; hover the row for a remove button. Each file is capped at 5 MB, since it's stored as part of the document's own saved content rather than uploaded anywhere — which also means it's included automatically in every existing backup path (local safety copy, auto-backup-to-file, export/import), with no separate attachment backup to think about
Documents saved before Files became its own tab still have their attachments as old-style inline chips inside the Notepad text — the first time such a document opens after updating, those chips are automatically lifted out into this tab and removed from the Notepad, with nothing to do manually
Version History doesn't keep the actual file bytes for either the old inline chips or this tab's list — same trade-off as pasted images — so restoring an older version never touches what's currently attached here, only the outline/Notepad/Q&A/etc. content that revision actually captured
Settings → Features → Pad Panel → "Files" toggle hides this tab entirely if you don't use it (on by default); Settings → Pad → "Default pad view" can also make Files the tab that greets you when the Pad opens

Remarks

One of the Pad's tabs — a record of things people said, worth keeping track of separately from the document's own content: each entry is a small card with who said it, the remark itself, a date, an optional short context line, and an optional link to a specific node in the outline
Click the + toolbar button to add a card — it's added at the top (newest first) and the Person field is focused immediately. Person/Date/Context commit on blur (click elsewhere, or press Enter); Escape while one of those is focused reverts that field's edit without touching the rest of the card
The remark text itself is a real editable field, not a plain text box — hover or focus the card for two small buttons above it: a bulleted list toggle, and Rewrite with AI (same one-field-at-a-time rewrite Decision Log's fields use, with an Undo toast after). Paste formatting works too, same as Notepad
The date defaults to today when a card is created but is freely editable — click it to open the same calendar popover Todo's due date uses (Today/Yesterday presets, plus month navigation for anything else), useful for logging something you're writing down after the fact, not just in the moment. The list itself sorts by this date (newest first), not by when the card was added — backdating a remark moves it to where it belongs
Linking a card to a node works the same as Files and Diagrams: a new remark auto-links to whichever node is currently selected in the outline; click the card's node label to change or clear that link via a searchable picker. A small accent-colored dot appears on a linked node — click it to jump straight to that remark (or, if a node has several, pick which one from a small list first)
Deleting a node that has a remark linked to it warns about the orphaned link first, same as it already does for Files/Diagrams/Decision Log; if a link does go stale (e.g. from restoring an older version of the document), the node label turns red and offers to re-link or clear it
Settings → Features → Pad Panel → "Remarks" toggle hides this tab entirely if you don't use it (on by default); Settings → Pad → "Default pad view" can also make Remarks the tab that greets you when the Pad opens
Right-click any node → Add remark… (on by default in the quick-action row, same place "Add question…" lives) creates a card already linked to that node and drops you straight onto its Person field, ready to type — no need to open the Pad first. Also reachable during a presentation, from Presenter Notes' own Remarks tab
Included in Word, PDF, and PowerPoint exports (Settings → Export → "Include Remarks in exports", on by default): a remark linked to a node appears right after it — same placement Decision Log uses — while an unlinked one lands in its own "Remarks" section at the end of the document (a trailing slide, in PowerPoint's case). Also reachable during a presentation: the Presenter Notes panel has its own Remarks tab, so a remark can be added or read without leaving Presenter Mode

Q&A

One of the Pad's tabs — a structured list: each row is a bold Question and a regular-weight Answer; hover either one for its own small Rewrite (AI) button, hover the row for its delete icon. A running "n/n answered" count sits in the toolbar. The scan icon scans the outline for nodes ending in "?", plus questions found inside notes attached to nodes, and offers them as a checklist; confirming adds them as new rows, and rescanning is safe since questions already added show disabled and unchecked instead of duplicating. It's selection-aware the same way the AI generate button is: a multi-selection scans just those rows, a single selected node scans that node's whole section, and nothing selected scans the whole document. Right next to it, the AI generate button (Settings → AI → Q&A AI → "Generate questions") reads the document and suggests questions worth asking about it — gaps, ambiguities, things it implies but doesn't actually answer — rather than only picking up nodes that literally end in "?"; it's selection-aware, so with a node (or a multi-selection of several) selected in the outline it scopes to just that section rather than the whole document, which also helps on longer documents where feeding everything at once dilutes what the AI focuses on. It opens the same review checklist as Scan, so nothing's added until you confirm, and these rows have no source-node link since they're new questions rather than ones already written into the outline. The Import button opens a paste box for bringing in Q&A from outside Sakura — it understands "Q:"/"A:" pairs (including Sakura's own Q&A text export, so re-importing an exported .md file round-trips cleanly) and falls back to treating every "?"-ending line as a standalone question if there are no Q/A markers at all; it opens that same review checklist too. The Copy button (next to Export as PDF) is Import's counterpart going the other direction — copies every question, answer, and section header as that same "Q:"/"A:"/"Section:" text, so pasting it into Import on another document brings the whole thing across cleanly; the bulk toolbar's own Copy button (next to bulk Rewrite, once rows are selected) does the same for just the selected rows instead of everything. The toolbar groups these three (Import/Scan/Generate) together as "bring in content from elsewhere," set apart from the + button and its neighbor — one blank row, one section/topic header — which are for writing your own from scratch. New rows and sections both land at the end of the list regardless of where you're looking, so drag the grip to move a row under the section it belongs to, or hover a section row for a small "+" that inserts a new blank question at the end of that section directly, without a drag. Sections drag and delete like any other row but have no select checkbox of their own — deliberately, since a group heading isn't something you'd bulk-act on the same way a batch of real rows is — and don't count toward the answered total, and are dropped from view (rather than left as an empty label) whenever a filter or search leaves nothing under them. Past the filter/search group, the whole-list AI actions run in the order a person would actually use them — Group into sections first (Settings → AI → Q&A AI → "Group into sections": reads every question and organizes them into a small number of named sections in one pass, replacing any existing grouping with a fresh one built from scratch, though every question keeps its answer, follow-up flag, and Answered by; only the ordering and section headers change, and Undo restores the previous arrangement in one step), then Draft answers, then Summary. A row added from a scan keeps a small link icon (hover the question) that jumps straight to that node in the outline; the icon disappears on its own if that node is later deleted, and when that happens an orange "n unlinked" chip appears next to the answered count — click it to show just those rows, the same way the answered count filters to unanswered ones (the two filters can combine). Each row also has a small chip (hover to reveal, stays visible once set) for flagging an answer as not yet trustworthy: click an unflagged row to mark it "Follow-up" (amber) for your own uncertain answers, or let AI-drafted answers flag themselves automatically as "AI draft" (accent-colored) — either way the chip only clears when you click it yourself, so editing the answer text doesn't silently mark it reviewed. A matching "n follow-up" chip in the toolbar filters to both flagged types together, same as the other two chips, and all three filters can combine. Once a row has an answer, a small "Answered by…" field and a date appear alongside the follow-up chip — hover-only until a name's actually been typed in, same reveal-on-hover treatment as the follow-up chip, so an empty one doesn't sit on screen for every single answered row; once it holds a name it stays visible like any other row data. The date is set automatically whenever the answer changes, but who answered is always typed in by hand, since it's frequently someone else (a colleague from business or IT) rather than whoever's typing it into Sakura; recently-used names show up as autocomplete suggestions so repeat entries are quick without ever being assumed. Selecting several rows (see below) adds a bulk "Set Answered by" button next to bulk Rewrite — one name, applied to every selected row that already has an answer (skipping any that don't, rather than refusing the whole batch), with its own Undo. Turn on the select/reorder toggle (checklist icon in the toolbar) to reveal each row's drag grip and select checkbox — drag the grip to reorder rows by hand, or click the "n/n answered" count to show only unanswered rows — the two don't mix: dragging is unavailable while that filter is active, since a hand-reorder inside a filtered view has no single sensible "real" position to save it at. Deleting a row — one at a time, or several at once — shows an Undo toast for a few seconds rather than deleting silently. With the toggle on, Shift-click a row's checkbox for a range, or use the checkbox above the list to select or deselect everything; once anything's selected, a delete icon appears next to that same header checkbox for a batch delete, also with Undo. Escape clears the current selection and Delete/Backspace removes it, both only while focus isn't inside a question or answer field, so they never interfere with normal typing. Since Q&A lives in the same Pad element that Presenter Notes relocates, it flips right along with it — press Q in Presenter mode to open Presenter Notes straight into the Q&A view (N for Notepad) and work through questions without leaving fullscreen. Settings → Panels → Pad has "Include Q&A in exports" (Word and PDF append it as a final section — bold question, regular-weight answer, unanswered rows marked "No answer provided" — turn off to keep Q&A in the Pad only). Settings → Panels → Pad → "Row actions" switches field icons and the delete button between revealing on hover (compact) and staying always visible — separate from the select/reorder toggle above, which is always opt-in from the toolbar rather than a Settings default, since reordering and bulk-selecting are occasional, not everyday actions

Decision Log

A structured decision record — Context, Decision, Rationale, Alternatives, and Impact fields, a Status pill (Proposed/Approved/Rejected) you click to cycle through, and an author/timestamp footer. Lives in its own tab in the Pad, right alongside Notepad, Q&A, and Diagrams — not inserted as an outline node, and each one is a separate record you link to a node rather than something attached directly to it
Click in the Decision Log tab's toolbar to create one — this is the only way to start a new one, so the tab is always the answer to "where do I add a decision log". If the currently-selected outline node doesn't already have one, it's linked there automatically; otherwise it's created unlinked and the node picker opens right away so you can choose a free node
The tab shows every decision log in the document as a folded row (status pill inline next to the linked node, a one-line snippet of Context or Decision, author/date) — click a row to expand it in place and edit the full card; click it again, or another row, to collapse it. Only one is expanded at a time. Turn on the drag-to-reorder toggle (grip icon in the toolbar) to reveal each row's drag handle and reorder rows manually — same explicit toolbar-toggle approach as To-Dos, Q&A, and Diagrams now all use, rather than the drag handle appearing on hover; it's disabled while the "N open" or author filter is narrowing the list, since a filtered view's row order doesn't map cleanly back onto the full list
One decision log per node — the node picker (click the linked-node line inside an expanded card) greys out and explains any node that already has one. Pick "Not linked to a node" to clear the link, freeing that node up
A small accent-colored dot appears next to a node with a linked decision log — click it to jump straight to the Decision Log tab with that entry expanded. The same icon appears next to a folded node's "+N hidden" badge when a decision log is linked somewhere inside its collapsed subtree
A node referenced by a Meeting Notes link or a To-Do's link shows its own small dot too — a calendar icon for meetings, a checklist icon for to-dos — click to jump straight there, or to a small list if more than one links to that same node. Same folded-branch propagation as the decision log dot above. If a meeting or to-do links to the whole document rather than one specific node, look for a small document-shaped chip next to the title, right by the document status chip — click it the same way. Only shows for the document you currently have open
Delete a decision log from its row (trash icon on hover) — asks for confirmation first (Settings → Editing → "Confirm before delete", same setting used for nodes, Q&A rows, and diagrams), then removes it. Undo with the toast's own Undo button. Deleting doesn't touch the linked node itself
Each field supports light rich text — type "1. " or "- " at the start of a line for a real list, Ctrl/Cmd+B/Ctrl/Cmd+I for bold/italic, no toolbar needed — plus a small AI-rewrite icon that appears on hover
Settings → Panels → Pad → Decision Log: "Ask who decided" prompts for a name each time a new one's created (on by default; the timestamp is always added regardless). "Color field labels" colors the Context/Decision/Rationale/etc. labels — in the editor and in the Decision Log card in Preview — so each field reads at a glance, and bolds them at the same time (the two aren't separate settings). Status is separate and always shows as a colored pill either way, since that's a status indicator rather than a fillable field. "Fields to include" hides whichever of the five fields you never fill in, right in the card. "Split large ones onto their own slide" (Presenter/PPTX) gives a decision log with a lot of content its own slide instead of squeezing it under the node's bullets, past a configurable character threshold
Turning Decision Log off (Settings → Features → Pad Panel) hides its tab, node dots, and Excel export — existing decision logs already in the document aren't deleted, just inaccessible until it's turned back on. Clicking a node's decision-log dot while the feature is off shows a reminder of where to re-enable it instead of silently doing nothing
Export menu → Decision Log → Excel (.xlsx) pulls every decision log across the current document into one spreadsheet (timestamp, author, linked node's text, and whichever fields were filled in) — fields left blank export as empty cells; an unlinked decision log's node text exports as blank

Diagrams

A real draw.io editor, embedded directly — not a static image. Lives in its own tab in the Pad, right alongside Notepad, Q&A, and Decision Log. Each document can hold any number of diagrams
Click in the Diagrams tab's toolbar to create one — it opens full-screen in draw.io immediately. Click any existing diagram's row in the tab's list to reopen it the same way; the row shows a live thumbnail once it's been saved at least once
Generate turns a selected node's subtree (or, with nothing selected, the whole document) into a rough draw.io diagram automatically — no AI involved, purely structural: boxes for nodes, arrows for parent/child relationships. Meant as a fast first draft to refine by hand afterward in draw.io, not a finished diagram. Caps out at 60 nodes/10 levels deep per generation to stay readable; pick a smaller subtree and generate separately if your outline is bigger than that.
Generate opens a review screen first — every node in scope, with its proposed shape (plain box, a specific shape like Decision/Datastore/External, or "Edge label"/"Passthrough" to fold it into the diagram rather than draw its own box), whether it's a titled Container, whether its children form a Sequence (chained, vs. fanned out independently), and Sequence direction (vertical/horizontal). Everything's pre-filled with a best guess from the outline's own structure — adjust anything that guessed wrong, select multiple rows for the bulk toolbar at the top, then Generate. Classification is remembered per diagram (not on the node, and never as a document tag — nothing here shows up as a tag chip anywhere else), so regenerating the same scope later reopens pre-filled with whatever you set last time instead of starting over.
Settings → Panels → Pad → Diagrams → "Skip review before generating" (off by default) makes Generate one click — it renders immediately using that same best-guess (or, on a regenerate, whatever was set last time) with no review screen in between. You can still refine the result afterward directly in draw.io either way.
Only changes saved inside draw.io itself are kept — its own Save button, or Ctrl/Cmd+S, while the editor is open. Closing the full-screen editor (the × button, or Esc) always asks you to confirm you've saved, since there's no way to force a save from outside draw.io's own UI
Multiple pages/tabs within one diagram: draw.io's own page tabs (bottom-left of the editor) work as usual, and every page's content is kept in full when you save and reopen. But the thumbnail, and the figure shown in Preview/PDF/Word/PowerPoint, can only ever show one page — whichever was active in the editor at the moment you last saved. (An earlier version of this tried exporting every page individually behind the scenes; that turned out to actively corrupt the live diagram — deleting a page from it — so it was reverted. Nothing unsafe is attempted here anymore.) A "N pages" badge appears on a diagram's row in this list as a reminder when this applies. If you want more than one page to actually appear in the document, create them as separate diagrams (the + button) and link each to its own node instead of using pages within one diagram
Give it a title in the field at the top, and link it to a node with the anchor button next to it ("Not linked to a node" clears the link). Linking is optional but is what makes a diagram show up anywhere outside this tab — an unlinked diagram only lives in the Diagrams list itself. The node link can also be changed right from a diagram's row in the list — click the "Under: …" (or "Not linked to a node") line, same picker either way — without needing to open the full-screen editor first
A linked diagram appears as a large figure directly under its node in Preview, PDF, Word, and PowerPoint exports — same treatment as a note image marked "Feature as diagram" elsewhere in Sakura, so there's no visible difference between the two workflows in the output. Settings → Export & print → "Include Diagrams in exports" controls the Word/PowerPoint case specifically (Preview and PDF always show it); click the figure in Preview to enlarge it
A small accent-colored square-cluster dot appears next to a node with a linked diagram — click it to jump straight to the Diagrams tab and open that diagram (or the list, if the node has more than one). The same icon appears next to a folded node's "+N hidden" badge when a diagram is linked somewhere inside its collapsed subtree
Delete a diagram from its row in the list (trash icon on hover) — asks for confirmation first (Settings → Editing → "Confirm before delete", same setting used for nodes and Q&A rows), then removes it and its node link entirely. Undo with Ctrl/Cmd+Z or the toast's own Undo button, either works, for 5 seconds — same window as every other delete-undo toast in Sakura. Deleting doesn't touch the node itself
The copy icon next to delete duplicates a diagram — same drawing, same status, same node link, titled "(original title) (copy)". Useful for variations on the same base drawing without starting over. A pure data copy, not a draw.io operation, so it's instant
Each diagram has a status — Draft, In Progress, Review, or Final — shown as a small colored pill on its row and in the full-screen editor's header. Click either one to cycle to the next status. Purely organizational — nothing else in Sakura reads it (it's not exported or shown in Preview), it's just a way to track where each diagram stands at a glance across a document with several
Two more chips can appear on a row: Unlinked, when a diagram has no node link yet (a reminder it's invisible in Preview and exports until you link it), and Modified since export, when it's changed since the last time anything was exported from Sakura. The latter is approximate — Sakura tracks one "last export" timestamp app-wide, not per document, so it can't tell whether it was specifically this diagram (or even this document) that got exported
One diagram is treated differently: the Whiteboard, opened via the pencil icon in Presenter mode's bar or its right-click menu (see the Presenter mode section) — a single persistent, deliberately unlinked diagram meant for sketching mid-talk rather than illustrating a specific node. It's pinned to the very top of this list regardless of sort mode or manual drag order, shown with an accent-colored "Whiteboard" badge instead of the "Unlinked" warning chip (it's supposed to stay unlinked, not a problem to fix), and has no anchor button of its own for that same reason. Everything else about it is an ordinary diagram — same draw.io editor, same save/autosave behavior, same duplicate/delete — created automatically the first time you open it
Each diagram has a private, one-line note — a reminder to yourself, e.g. "needs sign-off before this is final." Type directly into the small field under a row (or the matching one in the full-screen editor's header) — no popup, no separate edit step. Enter or clicking away saves it, Esc discards whatever was typed since it was last saved. Never shown in Preview, PDF, Word, PowerPoint, or anywhere outside this panel
Diagrams, Q&A, and Pad content are all included in Sakura's Version History now (More menu → Version History for the open document, or right-click a document in the file explorer → Version history) — restoring an older version restores all three alongside the outline text. A restored diagram's thumbnail is blank until it's reopened and saved once in draw.io, which regenerates it automatically; the diagram's actual content isn't affected, only the cached preview image. Pad content restores as text only — any file attachments or pasted images that were in the Pad at that point in time are gone for good, replaced with a small note of what was there, since keeping every version of a 5 MB attachment would make History unusably large. Restoring a version with older revisions recorded before this existed leaves diagrams/Q&A/Pad exactly as they currently are, rather than guessing or wiping them — but if one of those diagrams or decision logs is anchored to a node that only exists in the newer state you're restoring away from, it becomes unlinked from any node once the outline rolls back (the data itself isn't deleted — it still shows up in the Diagrams tab or Decision Log list, just without a node to hang off — the same "not linked to a node" state either already shows for a deliberately-detached one). The "Version restored" toast names how many diagrams/decision logs this happened to, if any, so it's never silent
Open or close the tab with Ctrl/Cmd + Shift + G, same "press again to close" behavior as the other Pad tabs' shortcuts
Turning Diagrams off (Settings → Features → Pad Panel) hides its tab, toolbar button, node dots, and keyboard shortcut — existing diagrams already in the document aren't deleted, just inaccessible (and left out of Preview/exports) until it's turned back on
Import — brings in an existing .drawio/.xml file: drag one onto the dropzone, or click it to browse. "Paste XML text instead" is there too for XML that's only sitting in your clipboard rather than saved as a file. Either way it's added to this list, linked to the selected node if it's free, with its thumbnail generated automatically. A small "Copy a prompt built from your outline" link is included for anyone who wants an AI assistant's help getting started — paste the prompt into whichever one you like, save or copy its reply, then bring that back in through Import
Generate — builds a rough flowchart directly from the outline instead of a blank canvas: a selected node's subtree, or the whole document if nothing's selected, becomes a real, editable draw.io diagram added straight to this list (capped at 60 nodes / 10 levels — a toast asks for a smaller section if you're over that). Layout, sizing, color, and the diagram's XML are always built by deterministic code — AI is only ever asked for a small, whitelisted classification of that same outline (never coordinates, never XML, never a restructured hierarchy), which the deterministic renderer then draws:
- Structure — boxes tile top-down following the outline's own hierarchy. A node whose children are all childless (a flat list of steps rather than genuine sub-branches) renders as a single vertical chain with sequential arrows instead of fanning out sideways — keeps a long flat step-list narrow rather than blowing out the canvas width, and reads as one continuous sequence rather than alternate paths. That whole chain (the node plus every one of its steps) also gets a background card drawn behind it, tinted to a lighter version of that group's own color rather than one fixed neutral — a pink actor group and a blue UI group each read as their own section, not both sharing the same generic card — so a labeled group of steps ("Webshop Layer" and everything under it, say) reads as one section on the page itself, its own text sitting directly on that card as the section's title rather than getting a redundant box of its own. If those steps are themselves explicitly numbered too ("1 · Simulate order", "2 · Calculate weight", ...), the chain lays out left to right in a row instead of the usual top-to-bottom column — same "these are ordered stages" signal, one level deeper. A node with a real mix of children (some with sub-detail, some without) still fans out normally — unless every one of those children is explicitly numbered ("1. Webshop Layer", "2. ERP Layer", ...), in which case they're read as ordered stages of one flow rather than alternate branches and stack top to bottom instead, each stage's own full subtree rendered normally beneath it, connected from wherever that stage's own content actually ends (not skipping over it) to the next stage's start. Tag that parent node horizontal (or row) to lay those stages out left to right instead — same idea as the numbered-leaf case above, just for stages that still have their own sub-detail, where structure alone can't tell "these are still parallel enough to sit in a row" apart from "these are sequential phases, stack them" (a top-level phase sequence and a same-shaped row of sub-steps look identical on paper). Each stage keeps its own subtree laid out normally beneath it either way — the tag only changes whether stages sit side by side or stacked. All-or-nothing: every sibling in the group has to be numbered for this to kick in, so one numbered aside among ordinary children doesn't pull the whole group into sequence mode by accident. Numbering style is flexible — "1.", "1)", "1:", "1 · ", "1 - " all count, since people number lists differently and the character itself isn't the meaningful part
- Color — a branch color cycles through 8 hues at every genuine fan-out in the tree, not just the top level, so a real branch point several levels deep (not just the outline's own top-level headings) still gets visually distinct paths with zero AI involved; a chain group's steps stay one uniform color on purpose. An explicit node marker (Confirmed/Issue/Parked/Follow-up/N-A — each its own hue) outranks that, and a node's first tag outranks plain branch color — a tag deterministically hashes to one of 8 reserved hues, so the same tag text always recolors the same way without any keyword list to set up first. All three carry down to untagged, unmarked descendants. Once AI has classified anything in this diagram (see below), a node it left unclassified switches from its branch color to a neutral gray instead — a branch hue has no legend entry to explain it, so once other colors on the page do mean something, an unexplained one would read as a mistake rather than "nothing special here"
- Box size — width grows with the (possibly-shortened) label up to a cap, and only grows tall enough for two wrapped lines if the label actually needs them, rather than every box defaulting to the same fixed size. A decision diamond gets extra width/height on top of that (a rhombus's angled edges leave much less usable interior than a box's), plus a much tighter label-length budget than every other shape, since even the extra sizing isn't enough room for a full sentence
- AI's job — one JSON call for the whole scope classifies which nodes are a UI/frontend step, a service call, middleware/integration, a backend process step, an actual persisted data store (kept distinct from an ordinary backend step — only a table/database/record gets the data-store cylinder), an external system, an actor, a genuine decision fork, a pure aside/caveat that isn't a step at all, or a branch that's explicitly out of scope of whatever's being described — each with its own fixed color and, where it helps, its own shape (diamond, cylinder, dashed, dotted). It can also flag a trigger node whose resulting outcomes are written as its siblings rather than its children (something the outline's own hierarchy has no way to connect on its own) and draw that connection as an extra edge, without changing the layout at all. A results-into-two-genuine-paths pattern is usually clearer expressed by nesting the outline itself that way in the first place, though — that's deterministic and doesn't depend on the model noticing anything, where an AI-suggested edge is inherently probabilistic and won't fire on every run. Every one of these is validated against the actual outline structure before it's trusted — a claimed decision only renders as a diamond if the node genuinely has 2+ children in the real tree, regardless of how the model read its own wording, and any classification outside a fixed whitelist is silently dropped. Genuinely fanning/parallel content (rare — most content that touches multiple systems is still a linear sequence, just one that hops between them) can additionally switch the whole diagram to a layered, swimlane-style layout instead. No AI key configured, a locked vault, a provider error, or a response that fails validation all fall back to the plain deterministic version above instead of blocking the diagram or leaving it half-classified — the Generate button spins for the duration of the call so a slower provider still has a visible in-progress state, and the toast afterward says whether AI classification actually applied this run or not
- Tags — a node tagged decision, database (or db/datastore), excluded, ui, service, middleware, backend, external, actor, or note (a few spelling variants work too — see the source for the full list) sets that node's box type directly, the same shapes AI classification can produce — worth having even when AI didn't run at all, or when AI's own guess should be overridden for a specific node. An explicit tag always wins over whatever AI separately classified that node as. decision only sticks if the node genuinely has 2+ children, same structural guard AI's own decision classification uses, so a stale tag can't produce a diamond with no actual fork under it
- Edge labels — a node tagged edge-label, placed as a sibling directly between two other nodes at the same outline level, contributes its own text as a label on the arrow connecting those two neighbors instead of becoming a box itself — useful for annotating a transition ("Passes: freight group & qty", "Returns: pricing values") the way a real system-integration diagram would. It never counts toward the 60-node cap and never gets its own layout slot; delete it (or remove the tag) to drop the label again
- Only categories actually present on the page get a legend entry, drawn to the top-right of the diagram — never a static list of every possible category
- Clicking Generate again for the exact same scope (same selected node, or "whole document" again) offers to regenerate that diagram in place rather than adding a duplicate — confirms first, since it discards anything changed since inside draw.io; declining leaves the existing one untouched
- Meant as a rough first draft to refine by hand afterward, not a finished diagram

Mind Map

Freeform brainstorming canvases, one per document — like Diagrams, a document can hold any number of named maps, listed in the Pad's Mind Map tab, each opening full-screen when clicked. A small + in that tab's toolbar creates a new one; rename a map by clicking its title there or in the open canvas's own header
A map starts genuinely blank — no forced starting node. Double-click empty canvas to drop a new idea; any idea with no parent is a root, shown in slightly bolder text, and there's no limit on how many roots a map has — it's a forest of independent trees, not one privileged "center" everything must connect back to. Double-click any node (or select it and press Tab) to add a connected branch; Enter adds a sibling next to the selected node — for a root, that's another independent root nearby. Double-click a node's own text, or select it and use the right-click menu, to rename it
Drag a node to reposition it — while dragging, it snaps to line up its center with any other node's, showing a dashed guide line, the same idea as alignment guides in most design tools. Drag empty canvas to pan; scroll (or the −/100%/+ controls) to zoom. Click selects a node; Ctrl/Cmd+click toggles a node in or out of the selection without starting a drag, and Ctrl/Cmd+A selects everything. Dragging any node that's part of a multi-selection moves the whole selection together. Delete/Backspace removes the selected node(s) and everything branching from them, confirming first if there's more than one idea involved — any node, including a root, can be deleted this way
Each root's children get their own color, automatically cycled and inherited by everything under them — so every tree on the canvas gets its own palette, not just one. Right-click any node (or a multi-selection) for a color row to set the color by hand — recoloring cascades to the whole subtree, and a multi-selection recolors every selected branch at once
Any node with children shows a small fold badge — click it to collapse the branch (shows a hidden-idea count) or expand it again; also available as "Collapse/Expand branch" in the right-click menu. Expand all / Collapse all (next to Tidy up) do the same for the whole map in one click. This is purely a view state, like the outline's own fold: it doesn't undo, and Generate Outline still walks the full tree regardless of what's currently collapsed
Double-click a connector line (or right-click it) to add a short label to it — e.g. "if approved" on the line between two ideas. Leave the text blank to remove the label again
Any idea can carry a short note, separate from its own text — right-click it and choose "Add note…" (or "Edit note" if it already has one); a small note badge appears in its corner, click it to view or edit without going through the right-click menu, and "Remove note" clears it
The same Confirmed / Issue / Parked / Follow-up markers the outline uses are available here too — right-click an idea for the marker row, click one to apply it (click the same one again to clear it), shown as a small glyph badge on the node. Unlike the outline, clicking a marker badge here doesn't filter the canvas — a spatial, searchable canvas doesn't need that the way a long flat list does
Expand with AI… (right-click an idea) turns it into 3-8 new branch ideas in one step — reuses the exact same AI call the outline's own "Expand with AI" makes, just landing as canvas nodes instead of outline rows, positioned automatically without disturbing anything else already on the canvas. Needs AI configured (Settings → AI Rewrite) and the AI feature on
The connections icon in the header (needs AI configured) looks across the whole map for meaningful relationships between ideas that aren't already connected — specifically ones from different branches, the kind easy to miss while heads-down on one part of the map. Anything the model suggests that turns out to already be a parent/child pair, an existing connection, or no longer exists by the time the request finishes is silently dropped rather than added — only genuinely new, still-valid connections become real dashed links, each with a short label
The search icon (or Ctrl/Cmd+F while a map is open) finds ideas by their text — matches are highlighted on the canvas rather than filtered into a list, since seeing a match in the context of its branch is more useful here than a flat result list would be. Enter / Shift+Enter (or the arrows next to the search box) step through matches, panning and briefly flashing each one in turn
Beyond the tree itself, two ideas that aren't parent/child can be cross-linked: Ctrl/Cmd+click to select exactly two nodes, then right-click either one and choose "Connect these two." The link is drawn as a dashed line (to tell it apart from the tree's branch lines) and can take its own label the same way; right-click it for "Edit label" or "Remove connection"
Ctrl/Cmd + Z undoes the last add, delete, rename, drag, recolor, connect, link, or Tidy up; Ctrl/Cmd + Shift + Z (or Ctrl/Cmd + Y) redoes it — its own undo history, separate from the document's
The Tidy up button (next to Reset view) recomputes a clean auto-layout in one click, remembering whichever style this map last used. Its small caret opens a picker for the three styles: Horizontal (the default — every root gets its own row, each generation a step further right), Vertical (same idea rotated — generations step downward instead), and Radial (each root's tree radiates outward from a center point, generations forming rings further out — reads better for a handful of branches off one core idea than a hierarchical process). Freeform dragging always stays available regardless of which layout Tidy last applied
Any idea can be linked to a specific node in this same document's outline: right-click it and choose "Link to document node…" to search and pick one, or "Add idea from document…" (also in the canvas toolbar and on right-click over empty canvas) to pull one in — a leaf document node becomes one new linked idea, and a node with sub-points brings its whole nested structure across as a matching branch, each level individually linked back to its own source line, not just the branch's top idea. Picking the same document node twice makes two separate linked branches rather than reusing one, if that's what you want. A linked idea shows a small connector badge in its corner; right-click it for "Jump to document node" (closes the map, selects and flashes that node in the outline) or "Unlink from document." The outline side shows the same link the other way too — a small dot appears on any node referenced by one or more mind map ideas, next to its note/diagram/decision-log dots if it has those; click it to jump into the right map (or pick from a short list if more than one idea references it)
Clear canvas resets the currently open map back to genuinely blank — confirms first if there's real content, and is itself undoable
The Export button opens a menu with three ways to get a map back out:
Export as image saves a PNG of the whole map (fit to its content, not just what's currently on screen) — reads the actual rendered colors straight off the canvas, so it always matches your current theme and any manual recoloring, and respects whatever's currently collapsed
Export to Diagrams converts the map into a draw.io diagram in this same document's Diagrams tab — the map already belongs to this document, so unlike Generate Outline below there's no new-document detour. It's a direct, deterministic conversion of the map's own layout and colors (root nodes keep their accent tint unless manually recolored) and connector labels — no AI involved, so unlike AI-generated diagrams elsewhere this can't come out malformed. Needs the Diagrams feature on (Settings → Pad Panel)
Generate Outline walks the whole map depth-first and creates a brand-new document from it — the document you're currently in (if any) is never touched or replaced
Presenter mode's M key (and its right-click menu entry) opens a persistent Scratchpad map instead of the named list above — one per document, created the first time it's used and reopened every time after, the same "always the same one, deliberately unlinked" role Whiteboard plays for sketching. It shows in the Mind Map tab's list too, pinned first with a Scratchpad badge, so anything jotted there mid-talk isn't lost or hard to find afterward
Settings → Pad Panel → Mind Map turns the whole tab off: hides it from Pad, the Presenter mode Scratchpad button and right-click entry, and every node's link dots and badges. Every map is preserved and comes back exactly as it was when re-enabled

To-Dos

An app-level task list — not tied to any document, node, or Pad. One shared list across your whole workspace
Open it from the Hub button in the app bar, or Ctrl/Cmd + Shift + T; the same shortcut closes it again, and it switches straight to this tab if the panel is already open on a different one
Floating and draggable by its header; drag the thin strip on its left edge to resize its width (remembered between sessions). The button expands it full-screen the same way Note/Pad/Code do (temporarily overriding your custom width, restored when you leave full-screen)
Two header buttons handle moving the list in and out of Sakura: Import loads a Sakura To-Dos file — a .json export from another Sakura session or shared by a colleague — adding its tasks to the end of your list with its own Undo toast; any task whose text matches one you already have is flagged (not skipped) in that toast, since text alone can't tell a real duplicate from two tasks that just happen to read the same. Export opens a small menu with two formats: PDF prints the full list — Open then Completed — regardless of any active priority/status/due filter, so a filtered view never silently shortens the printout; Sakura To-Dos saves the same full list as a shareable .json file for a colleague to Import the same way
Type a task and press Enter (or use the + button) to add it — the box grows as you type so long tasks stay readable. Shift+Enter inserts a line break instead of adding the task, both here and while editing an existing task's text
The clock icon next to Export opens Version History for the whole list — same idea as a document's Version History, kept separately since To-Dos isn't tied to any document. A version is saved automatically roughly every 10 minutes a change is made, or click "Save a version now" right before something risky. Restoring one replaces the current list (after saving a checkpoint of what was there first, so restoring is never itself destructive) — keeps the last 20 versions, app-wide
Checking a task off moves it straight to Completed, most-recently-finished first; unchecking sends it back to Open
The Open row shows how many open tasks match the current filter, with a magnifying-glass quick-find icon, sort and filter icons at its right edge. Sort covers Manual order, Priority, and Due date; filter narrows the list to one priority level (or tasks with none). Any icon turns accent-colored whenever it's off its default (or quick-find is open) so you can tell at a glance that something's active
Click the magnifying-glass icon to open quick-find and type to search task text, across both Open and Completed. Matches are highlighted and non-matching tasks are hidden entirely (not just dimmed), so the list actually shortens as you narrow it down — this composes with the priority filter, so a priority filter plus a search term shows only tasks matching both. Esc clears the search text first, then a second Esc (or clicking the icon again) closes it; the search always resets when you close the panel. Dragging to reorder still works correctly while filtered — a task dropped next to a visible neighbor lands there in the full list too, even if other hidden tasks sit between them. This searches task text only, live, within the panel — for a version that also opens the panel and jumps to the match from anywhere in the app, Quick Assist's todo: scope still does that
Turn on the select/reorder toggle (checklist icon in the header) to reveal each open task's drag handle, then drag it to reorder. Completed tasks aren't draggable — they're always ordered by completion time
Editing a task's text supports Ctrl/Cmd + Z / Ctrl/Cmd + Y to undo/redo that text edit — its own history, shared across all tasks (app-level, like the list itself) rather than per-task. This covers text edits only: adding, deleting, reordering, and completing tasks keep their existing in-panel Undo toasts instead
Hover a task for a spark AI Rewrite button and a delete button — both sit as an overlay rather than reserving space, so task text keeps the full row width when you're not hovering. Deleting shows an "Undo" toast, inside the panel itself, for a few seconds in case of a misclick — past that window, Version History (the clock icon) can still bring back a deleted task as part of restoring an earlier snapshot of the whole list
The Completed section header shows a Clear completed button whenever it has tasks (respecting the current priority filter) to remove them all in one action, with its own in-panel Undo toast — a batch counterpart to deleting one at a time. All To-Dos toasts (Undo prompts, export confirmation, AI rewrite errors) show inside the panel while it's open rather than the app's global toast, so acting on one never closes the panel
An overdue count badge can appear on the Hub button itself, in the app bar, whenever you have open, overdue tasks — visible without opening the panel. Toggle it off at Settings → To-Dos → "Overdue count badge" if you'd rather not see it
Priority, due date, and link chips sit below the task text. Priority cycles None → Low → Medium → High on click; due date shows "Today/Tomorrow" or the date, turning red once it's overdue. All three stay visible once set, and only reveal on hover while unset. Clicking the due date chip opens a small calendar popover with Today / Tomorrow / Next week presets up top for the common cases, month navigation for picking any other date, and a "Clear due date" option. Clicking the link chip opens a small popover with two tabs — Web link (an optional display-text field, a URL field, and Save; a bare domain or path is treated as an https:// link automatically) and Link to a node (a search box over your active documents — the same lookup Quick Assist and Global Search use — with results grouped into Documents, for linking the whole document, and Nodes, for one specific spot in it; leave the box empty to browse and pick a document without typing its title first), plus a Remove link option that clears whichever kind is set. A web-link chip shows your display text if you set one, or the link's hostname otherwise; a document-link chip shows the document's title with a plain document icon, and a node-link chip shows that node's own text with a small hierarchy icon, in place of the link-chain icon, so either reads as "jumps within Sakura" rather than "opens a website" — and both stay live, tracking the current title/text rather than a snapshot. Once any kind of link is attached, an open/jump button appears in the row's hover actions, next to Rewrite and Delete: for a web link it opens a new tab; for a document or node link it switches to that document if needed (scrolling to and briefly flashing the node, the same way backlinks do, when it's a node link specifically). If the node or document has since been deleted, the chip is shown with a dashed outline and clicking either it or the jump button gives a "That node no longer exists" (or "That document no longer exists") toast instead of failing silently
A task promoted from a Meeting Notes action item shows a small calendar-icon "from meeting" chip instead — separate from the link chip above, since a task can have both at once. Click it to jump back to the source meeting note; a small × on it removes just the reference, not the task. If that meeting note was later deleted, the chip switches to a dashed "missing" style rather than failing silently
A task can either repeat or hold sub-tasks — never both at once. Click the Repeat chip to cycle None → Daily → Weekdays → Weekly; checking off a recurring task doesn't complete it, it just advances the due date to the next occurrence (a toast confirms the new date), so recurring tasks never pile up in Completed. Once a repeat is set, a small repeat glyph also shows on the due-date chip, even before you've set a due date, so it's clear at a glance that a task recurs
Hover a task and click Subtask to open an inline field — Enter adds it (Escape cancels), building a small nested checklist under the task with its own checkboxes, an n/300 character counter while typing, and a delete button per item on hover. A progress badge (e.g. "2/4") plus a collapse arrow appear once a task has any sub-tasks; the parent task's own checkbox is disabled and completes automatically the moment every sub-task is checked — you can't check the parent by hand while it has sub-tasks. Setting a repeat on a task removes its sub-tasks' ability to be added (and vice versa, adding a sub-task clears any repeat)
Right-click a task for a context menu — mark complete/open, duplicate the task, copy its text, move it to the top or bottom of the list, expand/collapse its sub-tasks, or delete it. Right-clicking directly on the task's own text, or on a sub-task row, falls through to the normal browser menu instead, so copy/paste and spellcheck still work there
Settings → To-Dos: show or hide the overdue count badge; auto-expire completed tasks after a set number of days; cap how many completed tasks are kept (oldest drop first); choose whether row actions (priority/due/repeat/link chips, Rewrite/Subtask/Delete) stay always visible or only reveal on hover; and whether a task's first sub-task list starts collapsed by default instead of expanded
Typing @ directly in a task's text renders an inline chip: @name for an assignee, @date (ISO date, @today/@tomorrow, or a weekday like @fri) for a date reference, and @status (waiting/blocked/done/review) for status. A suggestion popup appears as you type, offering the date/status keywords plus any names you've already used as @name elsewhere — Arrow keys to navigate, Enter/Tab to accept, Escape to dismiss. These are colored text only — display convenience within the task text, not linked to the priority/due-date/link fields above
#tag renders as a colored tag chip — same autocomplete popup, suggesting tags you've already used. >remark renders everything from the > to the end of the line as a muted italic aside. Both are display only, same as @ tokens
Quick-find already searches the raw task text, so it matches #tags, @mentions, and remark text without any special handling — priority stays a separate structured field with its own dedicated filter
Quick Assist searches your tasks too — type a task's text, or scope it with todo: milk — and clicking a result opens the panel and flashes the match

Meeting Notes

An app-level meeting log — not tied to any document. One shared list across your whole workspace, same idea as To-Dos
Open it from the Hub button in the app bar, or Ctrl/Cmd + Shift + M; the same shortcut closes it again, and it switches straight to this tab if the panel is already open on a different one
Click to add a blank meeting note, or the layout icon next to it for New from template — Meeting Notes, 1:1, Daily Standup, Weekly Status, Project Kickoff, or Retrospective, each pre-filling the title/Agenda/Notes fields (and a starter action item, for the templates that call for one) so you're not starting from nothing. Either way it opens expanded and ready to type into immediately. Notes are listed newest first, click any row to expand it in place and edit it, click again (or click another row) to fold it back up
Each note has a title, date, and an optional time of day (shown on the row next to the date, e.g. "2:30 PM") — just a quick time reference, not a full scheduling system, so there's no separate end time or duration to fill in. Below that: attendee chips (type a name, press Enter or , to add one; Backspace on an empty field removes the last), an Agenda field and a Notes field (both light rich text — type "1. " or "- " at the start of a line for a real list, or use the bullet/numbered buttons next to each field's label; Ctrl/Cmd+B/Ctrl/Cmd+I for bold/italic, with their own undo/redo, plus a small AI-rewrite icon that appears on hover next to each field's label). Highlight any text in either field and a small formatting popover appears above it — Bold, Italic, Strikethrough, Highlight (a yellow inline highlight, click again on already-highlighted text to remove it), Link, and Rewrite (AI, rewrites just the highlighted phrase); Link and Rewrite have no other way in from these two fields' own fixed toolbar. Settings → Editing → "Selection formatting popover" turns this off if it's not wanted. Ctrl/Cmd + Space in the title, attendees, Agenda, Notes, or any action item field opens the same Quick Insert character-insert menu used everywhere else in the app (em dash, en dash, arrow, check mark, cross mark, middle dot, date/time). Below the fields sits a list of action items
Action items are simple checkboxes with editable text. The Promote icon next to one adds it as a real To-Do — the task text stays clean; instead, the meeting's date (if set) becomes the new task's due date, and a small "from meeting" chip appears on the To-Do row linking back to this note (click it to jump straight there; a small × removes just the link, not the task). Once promoted the icon turns into a checkmark (green) and can't be promoted again, so re-clicking never creates a duplicate
Linked nodes: click the link icon to search — results are grouped into Documents (matched by title) and Nodes (matched by text, with a highlighted snippet), same layout as To-Dos' own link search. Leave the search box empty to browse and pick a whole document without typing its title first. A node-link chip shows a small node icon, a document-link chip a document icon, so either reads as "jumps within Sakura" at a glance. Click an attached link to jump straight there; if the node or document has since been deleted, it's shown with a warning color and clicking it explains why instead of failing silently
Export a single meeting note as a PDF (download icon on hover, next to trash) — title, date/time/attendees, Agenda and Notes with their real formatting (bold, italic, lists, links), Action Items as a checklist (completed ones struck through), and Linked Nodes as a plain list. To-Dos' own list export and Q&A's own export produce a PDF the same way
The share icon on hover (left of the PDF download icon) saves that one meeting as a Sakura Meeting Notes .json file — send it to a colleague and they bring it in with the header's Import button below, the same "share/import between Sakura users" pattern To-Dos and whole documents also use
The panel header's two buttons handle the whole list at once, separately from any single note: Import opens a small menu — Calendar file (.ics) reads any standard calendar export the same way Settings → Data → Outlook Calendar's own .ics import does (checklist of events, dedup against already-imported ones by the file's event UID, flags any recurrence too complex to expand); Sakura Meeting Notes reads a .json file exported by another Sakura user (the whole-list export below, or a single shared note) — unlike the .ics path, this doesn't skip likely duplicates, it imports everything and flags matches (same title and date as a note you already have) in the confirmation toast, so you decide what to do with them. Export saves every dated meeting note as one .ics calendar file — undated notes are skipped, since there's nothing to put on a calendar without a date. Each event gets a 30-minute default length (meeting notes don't track an end time) and lists attendees in the description rather than as real calendar invitees, since Sakura doesn't collect attendee email addresses
Delete a note from its row (trash icon on hover) — with an in-panel "Undo" toast, same pattern as To-Dos and the document-level panels
The clock icon in the panel header opens Version history for the whole list — same idea as To-Dos' own, kept separately since Meeting Notes isn't tied to any document. A version is saved automatically roughly every 10 minutes a change is made, or click "Save a version now" right before something risky. Restoring one replaces the current list (after saving a checkpoint of what was there first) — keeps the last 20 versions, app-wide

Journal

An app-level daily journal — not tied to any document. One shared list across your whole workspace, same idea as Meeting Notes and To-Dos, but keyed to one entry per calendar date rather than any number of freely-dated items
Open it from the Hub button in the app bar, or Ctrl/Cmd + Shift + J; the same shortcut closes it again, and it switches straight to this tab if the panel is already open on a different one
Click for today's entry — it's created the moment you type something into it, not the moment you open it, so opening a fresh date and then navigating away without writing anything leaves no empty entry behind. Entries are listed newest first, each row showing its date, mood dot (if set), a one-line preview, and any tags; click a row to open it
Each entry has an optional mood (five presets — click again to clear), free-form tags (type one, press Enter or , to add it), and a rich-text body — type "1. " or "- " at the start of a line for a real list, or use the bullet/numbered buttons next to the field's label, plus a small AI-rewrite icon that appears on hover next to the label (rewrites the whole entry). Highlight any text in the body and a small formatting popover appears above it — Bold, Italic, Strikethrough, Highlight, Link, and Rewrite (AI, scoped to just the highlighted phrase). Settings → Editing → "Selection formatting popover" turns this off if it's not wanted
The calendar icon jumps to any date — Today/Yesterday/Tomorrow presets, or pick a day from the month grid; days with an existing entry show a small dot. Opening a past or future date this way works the same as opening today's — write something and it becomes a real entry
Export a single entry as a PDF (top of the entry, download icon) — date, mood, tags, and the body with its real formatting. The export icon in the panel header instead exports every entry, newest first, as one PDF
Delete an entry from its card (trash icon) — with a confirm dialog first, since there's no separate Undo toast for this one
The clock icon in the panel header opens Version history for the whole journal — same idea as Meeting Notes' own, kept separately since Journal isn't tied to any document. A version is saved automatically roughly every 10 minutes a change is made, or click "Save a version now" right before something risky. Restoring one replaces every current entry (after saving a checkpoint of what was there first) — keeps the last 20 versions, app-wide
Settings → Features → Journal turns the whole feature (dock button and panel) off

Recap

A curated look back at what actually got done — not a stats dashboard. Open it from the Hub button in the app bar (it's the last tab in that panel); it reads across everything: every document (not just the one currently open), To-Dos, Meeting Notes, and the Journal
Three tabs — Today, This Week, and Last Week — with the actual date range shown just underneath so it's clear what window you're looking at (weeks run Monday–Sunday)
Content is grouped by document (then Meeting Notes, then Journal), most recently active first — each group's own icon is shown if the document has one set. Within a group, items are short plain-language lines: a green check for something completed, a blue plus for something new, and a muted pencil for an edit — capped at a few lines each with a quiet "+N more" rather than dumping everything
Editing a node's Note counts as activity here too, same as editing its main text — so work that happened entirely in a note (not the outline line itself) still shows up. Decision Log entries, Diagrams, Q&A questions, and Mind Maps count too, each labeled by type (e.g. "Decision: …", "Diagram: …") within its document's group
Click any line to jump straight to it — a document item opens that document and scrolls to/flashes the node, a to-do opens To-Dos and highlights it, a meeting note opens Meeting Notes to that entry, and a journal line opens that day's entry. Click a document's own title (icon + name) to just switch to that document without picking a specific line
Summarize with AI (bottom of the panel) turns whatever the current tab is showing into a status-meeting-ready bullet list — grouped by document the same way the panel itself is, with a bullet (or a couple, for a genuinely busy group) per document/To-Dos/Meeting Notes/Journal, written with enough specifics to actually say out loud in a meeting rather than compressed into a bare label. Also picks up Decision Log, Diagram, Q&A, and Mind Map activity, same as the panel itself. Needs AI configured (Settings → AI Rewrite) and the AI feature on; Regenerate and Copy are both available in the summary that opens
Settings → Features → Recap turns the whole feature (dock button and panel, including the AI summary) off

Library

An app-level repository for quick-reference notes — not tied to any document. Same idea as Meeting Notes, To-Dos, and Journal, but built for things you want to find again later rather than a log of what happened: a saved link, a snippet you paste in often, a short how-to, a screenshot with context
Open it from the Hub button in the app bar. Entries are listed favorites-first, then newest-first by when they were last edited, each row showing its title, link (if set), a one-line preview of the note, and any tags; click a row to open it, or the button to start a new one. Every row uses the same book-and-lines icon regardless of content — it doesn't change based on whether a link is set
Each entry has a title, an optional URL with its own optional label (a plain-language name for the link, shown in place of the raw URL everywhere it appears once set — the list row, search results, both), free-form tags (type one, press Enter or , to add it), and a rich-text note — type "1. " or "- " at the start of a line for a real list, or use the bullet/numbered buttons next to the field's label. Paste an image directly into the note (a screenshot, or an image copied from elsewhere) and it's embedded inline, scaled to fit the panel width
In the list, a row's link (its label if set, otherwise the raw URL) is itself clickable — opens directly in a new tab without needing to open the entry first
Favorites: the star icon on a row (appears on hover, stays visible once set) or in an open entry's header marks it as a favorite — favorited entries always sort to the top of the list. The toolbar's own star button filters the list to favorites only, and combines with an active search or tag filter rather than replacing it
Tags in the list are clickable — clicking one filters the list to just that tag, shown as a dismissible chip next to the entry count (click the tag again, or the chip's ×, to clear it). Combines with the favorites-only filter and the search box the same way — all active filters narrow the list together, not one-at-a-time
A small AI-rewrite icon appears on hover next to the note's label — rewrites the whole note, preserving lists/bold/italic structure (pasted images are left alone either way). Highlight any text in the note and a small formatting popover appears above it instead — Bold, Italic, Strikethrough, Highlight, Link, and Rewrite (AI, scoped to just the highlighted phrase). Settings → Editing → "Selection formatting popover" turns this off if it's not wanted
The magnifying-glass icon searches across title, URL, link label, tags, and note content at once. If a search or filter combination matches nothing, the list explains which filters are active rather than just showing an empty panel
Delete an entry from its card (trash icon) or from its row in the list — with a confirm dialog first, since there's no separate Undo toast for this one
The clock icon in the panel header restores the most recent saved version of the whole Library — a version is saved automatically roughly every 10 minutes a change is made. Unlike Meeting Notes and Journal, there isn't yet a browsable list of every past version to pick from — just the single most recent one, with a confirmation first
Library entries are included in global search and Quick Assist — type library: or lib: before a query to search Library only
Settings → Features → Library turns the whole feature (dock button and panel) off

Feature activation

Settings → Features — on/off switches for the handful of capabilities most people actually turn off (Note, Code, Pad, To-Dos, Meetings, Journal, Recap, Library, Mind Map, AI, Tagging, Quick Assist). Panel-specific extras like Decision Log, Q&A, Diagrams, Focus, Backlinks, and Hover Toolbar have their own enable switch inside that panel's own Settings section instead — see each one's Help entry below for details. A small dot appears next to "Features" in the rail when any capability here is off, as a reminder
Note Panel — per-node rich-text notes. Off: the note toolbar button, right-click action, note dot on nodes, and the note panel itself are all hidden. Existing note content is preserved — turning it back on restores everything
Code Block — per-node plain-text/code content, on its own Code tab inside the same panel as Note. Off: the Code tab, its toolbar/right-click entries, and the </> code-dot badge on nodes are hidden. Existing code blocks are preserved
Pad Panel — a docked whole-document workspace with seven tabs (Notepad, Q&A, Decision Log, Diagrams, Mind Map, Files, Remarks), separate from per-node notes. Off: the floating Pad toggle button (bottom-right of the editor) and the entire Settings → Pad section are hidden, taking every tab with it. Each tab can also be turned off individually while leaving the rest of Pad on — see Settings → Features → Pad Panel. Content is preserved either way
To-Dos — the app-level task list, not tied to any document. Off: the To-Dos tab and its content are hidden from the Hub, and the Settings → To-Dos section is hidden. Existing tasks are preserved
Meetings — the app-level meeting log, not tied to any document. Off: the Meetings tab and its content are hidden from the Hub, and the Settings → Meetings section is hidden. Existing meeting notes are preserved
Journal — the app-level daily journal, one entry per calendar date. Off: the Journal tab and its content are hidden from the Hub. Existing entries are preserved
Recap — the curated Today/This Week/Last Week look-back across every document, To-Dos, Meetings, and Journal, with an optional AI bullet-summary. Off: the Recap tab and its content are hidden from the Hub
Mind Map — a per-document brainstorming canvas; one of Pad's tabs, not an app-level panel of its own. Off: the Mind Map tab in Pad, any currently-open mind map canvas, and its entry point in Presenter mode's Scratchpad are hidden. Existing maps are preserved
AI Capabilities — the Rewrite, Expand, Summarise, Generate outline, Suggest icon, Suggest tags, and Restructure text (Import menu) features, plus the AI actions inside Recap, To-Dos, and Q&A. Off: the AI toolbar group, all matching right-click items, the "Restructure text" entry in Import ▾, and every section under Settings → AI are hidden. Word import (Import ▾ → Word) still works while this is off, but loses its automatic AI fallback for documents that have no heading styles to extract — those import as a flat list instead. Requires a provider API key (Settings → AI → Provider) before anything will run — select a provider and the key hint shows where to get one
Tagging — #tag chips on nodes. Off: the tag toolbar button, right-click action, and tag chips on all nodes are hidden. Tags are still stored in the document data; turning the feature back on restores them
Quick Assist — the search box in the status bar (Ctrl/Cmd+K), which handles both plain-English commands and general search in one place. Off: the shortcut does nothing, the box disappears from the status bar, and any open dropdown closes immediately. Search-result behavior inside it (on/off, category filter, result cap) has its own dedicated section at Settings → Bars & Menus → Quick Assist
Turning a feature off hides every entry point for it (toolbar, right-click, Settings sub-section, Help sub-section). The underlying data is never deleted — toggling a feature is non-destructive in both directions. Feature states are saved alongside your other preferences and restored on the next launch

Start with structure. Refine later.

Settings