Presentations

A presentation is a slide deck your assistant builds for you, in a conversation, from a plain-language request. It renders right inside the chat, plays like a real deck (keyboard, touch, fullscreen, speaker notes), stays editable by asking for changes, and exports to PDF and PowerPoint that look exactly like the slides on screen.

Every deck you can see is listed under Work > Presentations in the sidebar — your own, the ones colleagues shared with you, and the ones published to the whole company. The Work group is where what your agents produce lives: their runs under Operations, their decks here, and every other file they output or you gave them under Files.

Create a deck

Ask any assistant that can generate presentations — the General Assistant, or an agent with the presentation tool enabled — for slides, in your own words:

Create a 10-slide all-hands deck for Q3: highlights, the three numbers that matter, the roadmap for Q4, and a closing slide with my contact details.

The assistant composes the slides, and the deck appears in the conversation as an interactive card. When several templates are available to you and your request does not name one, the assistant first asks which template to use (see Template and theme). You can also start from the Presentations page: New presentation lets you pick the template and the theme, then opens the chat with the beginning of a request already typed for you.

Brief it like a designer

A good brief names the audience, the one message to remember, and the numbers you want on screen. The assistant chooses the slide compositions — a two-column comparison, a wall of key figures, a before/after, a diagram with an explanation, a roadmap, a takeaways slide — and keeps each slide readable from the back of the room. Ask for a short outline first when the deck is long, then let it build the whole thing in one pass.

Start from the last edition

For a deck you make again and again — the quarterly update, the monthly report — ask the assistant to copy the last one: "copy Quarterly Review, Q3 2026 as Quarterly Review, Q4 2026, then update the numbers". The copy keeps the template, the theme, the slide size, the header and footer settings and every slide with its speaker notes, so the new edition starts from the approved layout rather than from a fresh draft, and a footer that shows the deck's title shows the new one. The copy is yours: it is shared with nobody, starts with no comments, and the original is not changed. Any deck you can open can be copied this way, one shared with you as Can view or published to the company included. The assistant then changes only what differs, slide by slide (see Change a deck).

Slides can carry text, bullet and check lists, callouts, key figures, tables, feature grids, timelines, process flows, numbered cards, quotes, images (a product screenshot can wear a browser window or a device frame on a glow in the brand colours), a challenge-and-answer pair (two stacked cards joined by an arrow), logo grids, native charts, org charts, and diagrams drawn from a text description (flowcharts, sequences, mind maps). The assistant can highlight a word or a phrase with bold, emphasis, code style, or a colored tone — for example an accent to point at the one thing that matters, or success / warning / error for a verdict.

Architecture and ecosystem diagrams

Ask for "an architecture diagram of the platform", "how the modules feed each other" or "the ecosystem around the product" and the assistant draws a native diagram in the deck's own palette and fonts: product logos from the template inside the boxes, labeled arrows between them ("Threat data", "Findings"), a highlighted core the rest runs on, a band across the top for a portal or a marketplace, a titled panel around the modules of a platform, and a legend. A box or a panel can also wear a colour of the template's palette — on the brand decks, the product colours: XTM One purple, OpenCTI blue, OpenAEV orange, OpenCRQ blue — as its border, a tint of its body, its icon and the chips inside it (ask for "each product in its own colour"); the colour follows the slide's theme, dark or light, and the PowerPoint exports carry it as shown. The arrows are routed so they never run through a box - when the two plain elbows between two boxes are both blocked, the arrow takes the free lane between the rows or the columns - an exchange in both directions runs as two parallel lines that do not cross, every arrow lands on the middle of a side (the plain ties from a highlighted core fan out onto the middle of each box's top edge), and a label never sits on a box or on another label: it slides along its arrow, or beside it, to a clear spot. The drawing is never blown up to fill the slide - a diagram on three columns keeps the same box and type sizes as one on five - and a diagram a little too tall for its place closes its rows up before anything is made smaller. A box that only gives context (a "ghost") can carry its icon inline before its label (icon_tile: false) so its label stays centred and two such boxes on one row line up. A panel can be headed by the platform's own logo from the template, in its top-left corner, with or without its title after it (ask for "the platform's logo on the panel"), and by a row of up to six small pills on the same line - the platform's capabilities, in its top-right corner by default, or right after the logo and title - which takes no room from the boxes inside the panel (ask for "the platform's capabilities in the panel's corner"); the pills are never wrapped onto a second line, and when a row is too long for its panel the layout check says so and names the panel. A logo or a row of capabilities can also sit bare inside a panel, with no box around it (frame: false); the arrows to it end on the logo or the words themselves. In the live deck the diagram is interactive: hover or focus a box to light up what it connects to, click it to read the detail the assistant wrote for it, Escape to release. Everything is vector — the PDF export stays sharp at any size.

Org charts

Ask for "the leadership team", "the engineering org with its headcount" or "the hiring plan for next year" and the assistant draws a native org chart: every seat says who it reports to, and the deck lays the tree out itself - one line of cards per level, every card of a level as wide and as tall as the others, each manager centred over their reports and joined to them by one line along the gap, and a secondary manager drawn as a dashed line routed around the cards. Each kind of seat looks the same on every slide: a person (their photo from the template's people photos, or their initials, the name on one line and the role under it), an open seat (dashed, in the warning colour, marked Open role, with its target date), a planned hire (in the accent colour, with the period it is planned for - "+3 Platform engineers" for three of them), a team drawn as one card with its headcount, a team drawn as a bubble whose area is its headcount, and an external seat (a contractor, a partner or an agency, muted with a dotted outline). Ask for "the headcount under each manager" and every manager carries the roll-up the chart computes from the tree - people count one, teams their headcount, open and planned seats apart ("14 · +3 open · +4 planned") - or today against the target ("14 → 47"). Other looks: left to right, a leadership wheel around the root, compact cards, everything as bubbles, or cards over bubbles (the people as cards, their teams as bubbles under them), one tinted zone per branch, and a legend of the kinds drawn. A large organisation folds: from a level you choose, a manager's reports become one card with their count, or only the first levels are drawn; when a level is wider than the slide its last level stacks into a column under one line, and a chart that would still need type too small to read is refused with a notice rather than shrunk - the layout check then tells the assistant to fold teams or draw them as bubbles. In the live deck, hover or focus a seat to read its whole team's headcount, its tags, its note and its dotted managers; click a folded team (or press Enter) to open it in place - the deck itself does not change; the arrow keys walk the tree. The PowerPoint export (editable shapes) carries the chart as native shapes - each seat a group of its card or circle, its words as text - and its lines as connectors, so it can be edited in PowerPoint; the PDF stays vector. In a video, an org chart is built level by level (see Video Studio).

Charts and dashboard slides

Give the assistant the numbers — "ARR by quarter for the last eight quarters", "the share of Community versus Enterprise", "trials against a target of 150" — and it draws a native chart in the deck's own palette and fonts: bar, stacked bar, line, area, pie, donut, a gauge against a target, a bar and line combined, a waterfall for a bridge, or a butterfly - two groups compared rung by rung, such as individual contributors and managers per seniority level, the first drawn to the left of a central spine and the second to the right, every number written as a plain size with no minus sign, a level one group does not have (a 0) left empty on its side - no bar and no number, the level names on the spine (or on the left axis), and each side's total under the legend if you ask for it; or a world map coloured by country (see "Maps by country" below). Values read the way a board expects (€16.5M, 86%, thousands separators in your language) and are written on the figure: a small upright label just above each bar of a grouped bar chart, as on a dashboard card — set smaller where it would touch its neighbour, and turned on its side only when even the smallest size would — and, on a stacked bar, inside each segment that has room for it, with the stack's total above - or, on a horizontal bar chart, just past the end of each bar, where the chart keeps room for the widest total so a 100% on a chart of shares is never cut by its frame. A segment too thin for its number - a series worth a few percent of each stack - writes it just past the stack instead, after the total: a small figure in the series' colour, tied to its segment by a short line, several of them stacked neatly when one bar has more than one; the PowerPoint export draws them in the same place. Ask for smaller or larger chart text ("label size sm", as on a Metabase card; "lg" for a figure read from the back of the room): the values and the axis text follow, in the viewer and in the exports. To size the numbers apart from the axis text, ask for a value size - one of those text sizes, or a size in pixels from 10 to 32 ("values at 20px", two under the slide's 22) - and only the numbers on the bars, points and stacks change, set smaller from there if they must; the axis text keeps its size, in the PowerPoint export too. In the editor this is Value text size on the chart. When the values of a bar, stacked bar or combined chart are wider than the room between two bars - two series over seven quarters in half a slide - you choose what they do instead of being set smaller or turned: ask for them on two rows ("stagger the values") and every value keeps its size, upright, just above its bar - only a value that would touch the one beside it moves a row higher, and on a chart that combines bars and a line, a value the line would run through moves above the line, so no number is crossed out by it; or ask to keep them upright, and every value keeps its size in its place: a value that would touch its neighbour or the line moves just one row higher when that row is free, without moving any bar, and one that would still touch something there is left out, which the layout check reports. Either way the bars stand a little further apart, or narrow a little, to make room - except the bars of a grouped bar chart, which keep the tight groups they always have, so each period still reads as one group: there a value wider than its bar runs over the gap and the bar beside it, its halo keeping it legible on that bar, and only another value moves it a row higher or leaves it out. The PowerPoint export spaces its bars the same way. A horizontal bar or stacked bar chart takes the same choice: staggered, every value keeps its text size - one that would touch the value in the next row moves along its row just past it, and a segment's number too long for its segment is written past the stack, after the total, at the same size and tied to its segment by a short line - so two share charts on one slide read at one size; kept upright, a value that would touch another, or that its segment cannot hold, is left out, which the layout check reports. In the editor this is Value labels on the chart. On a pie or a donut each value sits just outside its slice, at the end of a short line, and the pie leaves exactly the room its values need, so a value is never cut by the card or written over the legend: in a narrow panel the pie is drawn a little smaller, and the values are set smaller only when the pie would otherwise shrink too far. When even the smallest size leaves no room outside - a pie squeezed beside a long legend - the values are written inside the slices large enough to hold them, in a colour that reads on each slice, and a slice too small keeps its number in the tooltip. The PowerPoint export writes the values at the same size and on the same side of each slice. Values turned on their side are reported by the layout check (values rotated on chart 2) so the assistant can put them on two rows, pick a smaller text size or a wider column; so are a pie's values set smaller than its text size (values shrunk on chart 2), so the assistant can move the legend under the chart or give it a wider column. Should a chart have truly no room for its numbers, the layout check says so (values hidden on chart 2) so the assistant gives it more room or fewer series; values are never dropped without a word. For a stacked chart it only suggests a taller or wider chart when that would actually bring the numbers back; when no size would, it says so and suggests leaving that series' numbers to the tooltip, or leaving the chart as it is - the totals are written either way. Figures are as precise as they need to be: a value under 10 keeps two significant digits (0.96x, 1.4h), and two different values never read the same (1.57x and 1.56x, not 1.5x twice). When a slide states its figures to a given precision, ask for it - "two decimals on the burn multiple" - and every value read against that axis, its ticks and tooltips included, is written with exactly that many decimals (1.27x, €19.14M), in the exports too. By default each amount takes its own unit, so one chart can read €480k next to €1.2M; ask for one unit - "all in millions", or thousands, or billions - and every amount on the chart, its ticks, tooltips and target included, is written in it with the chart's decimals (€0.25M next to €2.50M, an amount too small for the unit reading €0.00M rather than disappearing), in the exports too. Pies, donuts and gauges keep their own amounts. A gauge stands in the middle of its card or dashboard panel, its dial filling the start of its track and rounded at the value; ask for a small, medium or large one ("size sm", md by default, lg: a dial 240, 360 or 480 px across, in a dashboard panel too), or for one that fills its room ("fill"), and the dial grows as far as the room allows without cutting its number or its target label - in the PowerPoint export too, where the gauge is a picture of the figure. A line drawn over bars writes its values as well - each next to its point, over it or under it, on the bar with a white halo where the point falls in one, never on an axis label or one number over another: a value set too close to another moves just clear of it, and a line's first value slides clear of the axis beside it. A line or an area whose values would run into one another in a single row - eight quarters in a one-third dashboard panel - writes them alternately above and below the points, at the largest size that gives each value two points' room, before it sets them smaller or leaves any out; the PowerPoint export places them the same way. A chart where nothing touches is drawn exactly as before. To part a line from the bars on purpose, ask for an axis range - "make the left axis go up to €60M", "start the right axis at -€5M" - or set Axis maximum / Right axis minimum in the editor; the range always keeps zero and every value of its axis. To leave one series' numbers to the tooltip, ask for it ("no values on the burn bars", or on the small series of a stacked chart, whose stack totals still count it). A target shows as a dashed reference line named in the legend, and the chart follows the theme when you switch it between dark and light. The legend sits above the plot - on the panel's title row in a dashboard, beside a chart's own title - unless you say otherwise: ask for "the legend under the chart" (or set Legend position to Bottom in the editor) when a narrow panel's entries would otherwise wrap beside the title, and it moves to one centred line under the axis with the plot taking the room; Right stacks the entries beside the plot, the natural place for a donut's slices. The PowerPoint export puts the legend on the same side. A chart stands on a card — the deck's card surface and a fine border, like the key figures beside it and the panels of a dashboard — with its legend and caption inside; ask for a chart "without its card" (or turn Framed off in the editor) to draw it straight on the slide. Give a chart a title ("title it Monthly signups", or the Title field in the editor) and it is written at the top of the card, as a dashboard panel titles its chart, with the legend on the same row; the caption stays the note under the plot. Two by two charts on a slide, each titled and asked to fill its space, take their cards to the bottom; a chart in a dashboard panel is titled by the panel, which refuses a second title. The card is there from the first frame and stays while the bars grow in, and the editable PowerPoint carries it as a plain shape under the chart. Two units can share one chart — new logos as bars next to the average sales price as a line, or burn in millions next to the burn multiple: the second series reads against an axis on the right, written in its own unit (€300k, 1.5x) and drawn in that series' colour, with a target line of its own if you give one. In the live deck a chart is interactive: hover a bar or a point for its exact value, click a legend entry to hide or show a series. A chart asked to fill its space — on its own, under a heading in one of two columns, or inside a card that fills its column — takes the height left once the blocks around it are placed, and never less than a readable plot: when there is not even that, the layout check reports the chart as too short, so the assistant trims what surrounds it. A waterfall shows a bridge — "the ARR bridge from Q2 to Q3: opening, new logos, expansion, churn, ending": the opening and ending stand on the axis, each change floats from where the previous one left the total — rises in green, falls in red — and dashed lines carry the level from bar to bar. The platform adds the bars up: an ending that does not equal the opening plus the changes (beyond half a percent of what the changes moved, the slack of rounded figures) is refused with the gap named — "ending €19.14M vs running sum €19.2M, gap €0.06M" — so the assistant corrects the figure that is wrong. When the gap is real, ask for it to be shown: it is drawn as its own Unexplained bar before the total instead of being hidden in it. For a slide that reproduces a dashboard, ask for a dashboard slide: a row of key figures across the top — each can carry a small trend line beside its value — and up to six panels of charts, tables or figures on a grid below, sized to fit the slide without shrinking. Two panels can also share the grid half and half - a chart beside a table: ask for "two equal columns" (in the editor, Panel columns set to 2 on the slide) and they sit side by side at the same width, up to four panels on the slide; the PDF and PowerPoint exports draw the same two columns. When five or six key figures each carry a long comparison under them ("Q3 25 EUR 4.5M · Q2 EUR 6.7M"), ask for small tiles (or set Tile size to Small on the slide in the editor): the figures and their notes are set smaller so each note stays on one line, and the panels below get the room back, in the PDF and PowerPoint exports too. A row of key figures on a content slide comes in three sizes — small, medium (the default) and large — and a size reads the same on every slide: when a slide holds more than fits and is shrunk to fit, what surrounds the row shrinks, not the figures, so a KPI row repeated across a deck keeps the same size from one slide to the next, in the PDF and PowerPoint exports too. Only a slide too full to fit even that way shrinks its figures as well, and the layout check says so. A key figure's change is green when it rises and red when it falls, unless you say which way is good news: when lower is better - a loss, a burn, churn, costs - the assistant shows a fall in green with its arrow still pointing down ("-18.4% QoQ"), and a rise in red. In the editor, Delta tone sets that colour on a figure by hand, and the PDF and both PowerPoint exports show the same colours.

Maps by country

Ask for "customers by country" or "headcount per country" and the assistant draws a world map (a choropleth): each country with a figure is coloured on a ramp of the template's chart colour, from a light wash for the lowest figure to the full colour for the highest, and the countries without one stay in the muted surface colour, in the dark and the light theme alike. The countries are named by their two-letter ISO 3166-1 code (FR, US, GB); a code that is not one — UK for the United Kingdom, a country name — is refused with its position named, so the assistant corrects it before the deck is saved. Hover a country to read its name, in your language, and its figure; with values shown, the figure is also written inside every country large enough to hold it, as large as the country allows, and a country too small to see (Singapore, Malta) gets a dot in its colour. The legend is the ramp itself, with the lowest and the highest figure. Ask for a stepped legend instead and the figures are grouped into at most five classes, cut where they naturally part (or into the number of classes you ask for, from two to seven): each class is one flat shade of the same colour, every country takes the shade of its class, and the legend lists the classes beside the map, a dot and a range each — 43 for a class of one figure, 12 - 39 for a range, 120 + for the top class. Ask for no dots and the small countries keep their colour only, however tiny that is at the map's scale; a territory with no shape of its own on the map (Réunion, French Guiana) keeps its dot, so its figure is never lost. In the editor, the Map legend, Map classes and Small-country markers controls set the same. To zoom, ask for the map of Europe, the Americas or Asia-Pacific: the neighbouring countries are still drawn at the edges, and the layout check names a country that falls outside the region. A map takes a title, a caption, a frame, the full height of its slide and the text sizes like any chart, on a content slide or in a dashboard panel. In the PDF it is vector, like every chart; in the editable PowerPoint it is a high-resolution picture of the map, its title, caption and legend written as text beside it — PowerPoint's own map chart only opens in Office with an internet connection, so it is not used.

Agenda slides

A short agenda is a numbered list — one column up to four items, two columns up to six. From seven items the assistant lays it out as a board-deck grid: three columns of N. | TITLE cells, each with a one-line description under the title, so a long agenda still reads from the back of the room. The grid holds six to nine items — two to three full rows — and past nine the agenda returns to a two-column list rather than shrinking. Ask for "the agenda as a grid" or "as a list" to choose the look yourself, or "in three columns" to get the grid from six items; a description on any one item reserves that line on every cell, so the board stays even.

Logo walls

Ask for "our customers by sector" or "a logo wall of our partners" and the assistant groups the logos into equal cards — one per sector or category, its title and icon on one line — that fill the slide, each card's logos spread across its full width. Every logo is sized by its visible mark, not by its file: the transparent margin a logo was exported with is cropped away, so "small", "medium", "large" and "extra large" (28 / 40 / 56 / 72 px on the slide) are the height you actually see, and a round seal is drawn a little taller than a long wordmark so both carry the same weight. A logo row that is too narrow shrinks all its logos together, never one alone. Ask for the logos in one colour (the slide's text colour, or a muted grey) for a quiet "trusted by" band — it is also how white brand logos stay visible on a light slide; hovering a logo in the live deck shows its original colours. Labels under the logos are off unless you ask for them. Boxes of text follow the same idea: ask for "a card" instead of a tinted panel, with the icon on the title line and the content using the full width.

Customer quotes

Ask for "a customer quote slide" or "the customer's quote as a statement" and the assistant makes the quote the slide itself: large and bold, centred on the slide, with an oversized quote mark in the deck's accent colour, the speaker's name, role and company underneath and, when you give them, two to four facts about the customer ("Employees 80K", "Revenue $30B") as figures in the accent colour. A quote can also sit beside other content as a pull quote with a large accent rule, or be a testimonial card with the speaker's portrait or the company logo. In the editor, a selected quote offers its four looks as thumbnails, the size (up to the size of a title-slide headline), the alignment, the quote marks, the portrait and the logo picked from the deck's template, and the figures. Key figures anywhere on a slide can drop their card the same way (the plain variant of a metric), so a row of numbers sits under a paragraph without turning into boxes.

Funding rounds and partner logos

Ask for "our funding rounds with the lead investors' logos" and the assistant draws a timeline whose steps each carry their investor's logo, under the step's description and centred on its step, all the logos on one row. In the editor, each step of a timeline has a logo picker over the template's logos. A logo grid with a set number of columns shares its width in equal cells, each logo centred in its own, so a grid placed under a timeline of as many steps lines up with it; Stretch spreads the logos edge to edge instead.

Sales and marketing collateral has its own building blocks: a feature matrix (a comparison table with an accented "us" column and check / cross marks), editions and plans as tier cards with one recommended tier and a call-to-action button each, a fact sheet made of badges (certifications, hosting regions, community numbers) and label / value rows, personas — two to four audience cards naming the role, what hurts today and what changes for them — and a banner for a hero header or a closing call to action with buttons that link to a page or an e-mail address. Ask for "a battlecard against a legacy TIP", "the editions slide", "who it is for" or "a closing slide with a demo button" and the assistant reaches for them.

Pages, not only slides

Not everything is presented from a stage. Ask for a product brief, a one-pager, a datasheet or a case study and the assistant lays it out on a portrait page — A4 or US Letter — instead of a 16:9 slide: a hero banner, a fact sheet, a feature matrix, the editions, and a call to action, denser than a slide because it is read, not projected. A page exports to a PDF on real A4 / Letter paper and to a portrait PowerPoint of the same size. On screen it is not a slide show — see Read a page.

Images come from the predefined assets of the deck's template rather than from the assistant's imagination: add your architecture diagram, product screenshots or partner logos to a template, ask for "the architecture diagram" or "a logo wall of our partners", and the assistant places the actual figure — every one is described so it can be picked by what it shows, and the assistant asks the template for one role or keyword at a time rather than reading the whole inventory. See Templates.

Play a deck

The card in the chat is a live deck. Everything you operate it with sits in the toolbar along its top edge — nothing is drawn over the slide itself: no corner arrows, no progress bar, no counter on the slide.

The middle of the bar is navigation: previous / next, and the slide counter (3 / 12), which opens a go-to-slide menu listing every slide by title, with a number field for long decks.
Present — the one highlighted button on the bar — takes the deck, toolbar included, to the whole screen. The slide fills the screen edge to edge: a 16:9 deck on a 16:9 monitor has no border at all, and on a screen of another shape (a 16:10 laptop, an ultrawide) the bands around it are black, like any player, whatever theme the deck or the app is in. The bar floats over the top edge for a moment, then slides away with the pointer after a moment of stillness and comes back as soon as you move the pointer — never when you press a key, so stepping through the slides with the arrows or a clicker never makes it flash; press Esc (or the bar's exit button) to leave. The keyboard is always active while presenting. Where the browser will not hand over the screen — a locked-down kiosk, an embedded desktop window — the deck fills the application window instead, with the same bar and the same keys; and from a chat embedded in another product, Present opens the standalone viewer, which can go fullscreen.
Open in studio, on the card in the chat, lands on the standalone viewer at /presentations/<id> — the same page a deck opens on from the Presentations list, with the editor, the comments, the history and the sharing controls — and opens the studio beside it, carrying this very conversation along so the assistant keeps everything it already knows about the deck. It is also the page you put on a projector or share as a link: no product menus, only the deck's own toolbar. Open in a new tab, in the More menu, opens that page beside the chat.
Comments opens the review notes on the deck and History its revisions (see Work on a deck together); the count on Comments is the number of notes still open. The initials beside Present are whoever else has the deck open right now.
The More menu holds the rarer presenter actions: the slide overview, the black screen, the speaker view with your notes and the next slide, the printable notes handout, and the list of keyboard shortcuts.

In the chat, the keyboard belongs to the page until you click the deck or its toolbar; from then on the deck has the keys, until you click somewhere else. In the viewer and while presenting the deck always has them.

Keys	What they do
`Right` `Space` `N` `Page Down` / `Left` `P` `Page Up`	Next / previous slide
`Home` / `End`	First / last slide
`G`	Go to a slide
`O`	Slide overview
`F` / `Esc`	Present / stop presenting (`Esc` also closes the overview)
`B` or `.`	Black screen
`S`	Speaker view
`?` or `F1`	Keyboard shortcuts

Shortcuts pause while you type in a field or use a menu, so a key never turns the page by accident.

The deck is authored at a fixed 16:9 slide size (or 4:3 if you asked for it) and scales to whatever space it is given — a chat card, a laptop screen, a projector — so what you rehearse is what the room sees.

Present with your notes

Every slide can carry speaker notes — what you say over it, not what the room reads. The assistant writes them with the deck (ask for "speaker notes on every slide, conversational, two or three sentences each" and it does; the note under each slide in the editor panel shows the words and the speaking time it adds up to), and reports back which content slides still have none. They live in the deck, travel with it in the PowerPoint export, and are read from the speaker view.

S, or Speaker view in the More menu, opens the speaker view in a window of its own — from the standalone viewer or from the card in the chat alike. Put the deck on the room's screen and keep this window on your laptop:

The current slide on top, so you know what the room sees without turning round — with a "screen is black" notice while the deck is paused.
Your notes in large type under it, the reason the window exists: bullets as bullets, emphasis as emphasis. + / - set the type size once and it is remembered.
The next slide in the side column, so the transition is spoken before it appears.
A stopwatch that starts on your first slide change (pause, resume, reset at will), the wall clock, and your pacing: the notes give every slide an estimated speaking time, so the window knows how far ahead of or behind the plan you are, and how much spoken time is left. Type a target length in minutes and the plan is re-scaled to the slot you actually have.
The transport: previous / next, a go-to-slide menu, the black screen — every one with the same key as the deck.

The window follows the deck: whichever screen is playing it — the viewer, a chat card, the fullscreen stage — announces where it is, and the keys pressed in the speaker window move the room's screen. Your clicker moves both. With no deck window open it runs in rehearsal mode, and says so: same notes, same timer, its own position — the way to practise the talk before the room is there. Edits made from the studio while you rehearse land in the notes and the previews without a reload.

Notes handout, in the same menu (and in the speaker window's header), opens every slide with its notes as one printable document at /presentations/<id>/notes — the sheets for the lectern, or for the colleague who presents the deck in your place. Only slides with notes leaves the section dividers out; print it to paper or to PDF from the browser.

If the speaker window does not open

Browsers block windows a page opens without a click behind them. Allow pop-ups for this site and press S again — the speaker view is a new window, never a popup drawn over the deck.

Read a page

A portrait presentation — a brief, a one-pager, a datasheet on A4 or Letter — is a document, and the platform treats it as one. Instead of a stage it opens as a reader: every page stacked on a scrolling surface as a sheet of paper, the same toolbar along the top edge, and nothing that only makes sense for a slide show — no fullscreen, no transitions, no speaker view.

The middle of the bar is navigation: previous / next page, and the page counter (1 / 2), which opens a go-to-page menu listing every page by its heading. Scrolling moves the counter too — the current page is the one under the top of the reader.
The zoom controls sit beside it: zoom in / out in steps, and Fit width, which sizes each sheet to the reader. Ctrl + mouse wheel zooms as well.
Open in studio, on the card in the chat, lands on the standalone reader at /presentations/<id> with the studio open beside it and the document's actions (edit, comments, history, share, export) alongside; Open in a new tab in the More menu opens it beside the chat.

Keys	What they do
`Right` `N` `Page Down` / `Left` `P` `Page Up`	Next / previous page
`Home` / `End`	First / last page
`G`	Go to a page
`+` / `-` / `0`	Zoom in / zoom out / fit width
`?` or `F1`	Keyboard shortcuts

What you see is what prints: each sheet on screen is the page the PDF export produces, scaled as a whole, so a line that fits on screen fits on paper. In the library, pages carry a paper badge (A4, LETTER) and open with Open rather than Present; the Decks / Pages pills above the list show one kind at a time.

Change a deck

You never rebuild a deck from scratch. Ask the assistant for the change in the same conversation, and it changes the deck in place: the card in the chat and the link you already shared show the new version, and the viewer keeps your current slide instead of jumping back to the first one.

Move the roadmap before the numbers, add a slide comparing the two options, and make the closing slide warmer.

A small change stays small. "Change revenue to ARR on slide 12", "shorten this title", "swap the two columns", "delete the second bullet" edits exactly that slide, block or word: the assistant applies the change to the deck as it is stored rather than writing the whole deck again, so the other slides are not redrawn, reworded or reshuffled along the way — on a forty-slide deck as much as on a five-slide one. The same holds inside a dashboard slide: "stagger the values of the ARR chart on slide 7", "rename that panel", "widen the trend chart" or "drop the region table" changes that one panel, and the other charts, tables and key figures of the dashboard are not written again. The reply says which slides changed and confirms that the rest are untouched, and the deck's History records the same thing (Edited slide 12). The layout check that follows every save looks at the slide that was edited — if the new wording no longer fits, the assistant trims that slide, not the others — while anything that was already tight elsewhere is left for you to raise, or for the viewer's Layout check; a change to the whole deck (a new theme, another aspect ratio) is checked on every slide, since every slide was redrawn. When two people edit the same deck, the second edit is applied to the deck as the first one left it — and if slides were added or reordered in between, the assistant re-reads the deck before touching a slide by its number, rather than editing whatever slide now sits there. Only a request that is genuinely a rewrite — a new outline, a different story — regenerates the deck.

Point at the slide

Naming the slide ("on slide 12", "the closing slide", or, in the studio, "this slide") is what keeps an edit surgical. A word that appears on several slides is changed only where you pointed; without a pointer, the assistant is shown every slide the word appears on before anything changes, so it changes them all on purpose or asks you which one you meant.

Small fixes need no assistant round-trip. In the standalone viewer, Edit slides turns the deck itself into an editor, the way a slide editor is expected to work, with a side panel beside it.

Edit slides by hand

On the slide. Click a text and type: the caret lands where you clicked, the slide keeps its look while you type, and the panel's fields follow. Click a block — a list, a callout, a card, a table — to select it: an outline shows what you have, a small bar above it carries its actions (move up or down, duplicate, insert a block after it, delete), and the Selection tab of the panel shows every setting the block has. Clicking a selected list row, tile or card opens the one text it holds; a double-click does both at once; clicking the empty part of a slide selects the slide.

While a text is open, the bar above it formats the selection or the word at the caret: bold, emphasis, code, a tone (accent, info, success, warning, error, muted, gradient), clear formatting, a line break. What you see is what the deck stores — a formatting the deck's grammar cannot carry (an emphasis glued to the middle of a word, say) is written as plain text rather than as broken markup.

From the keyboard, with a block selected: Enter edits its main text, Tab / Shift+Tab move the selection to the next / previous block of the slide, Del deletes it, Ctrl+D duplicates it, Alt+Up / Alt+Down move it, Ctrl+S saves at once (everything saves on its own anyway), Esc leaves the text, then the selection, then hands the keys back to the deck. While typing in a text: Enter inserts a line break, Ctrl+Enter (or a click elsewhere) closes the text, Esc takes back what you typed in it, Ctrl+B / Ctrl+I / Ctrl+E toggle bold / emphasis / code. You never need to close a text to keep it: what you type is saved while you type.

Undo and redo — Ctrl+Z and Ctrl+Y (or Ctrl+Shift+Z; Cmd on a Mac) — work from anywhere while the panel is open: on the slide, on a floating bar, in the panel and in its fields, so a word typed in the inspector is taken back with the same keys as a block moved on the stage (the Undo / Redo buttons in the panel footer do the same, and the shortcuts sheet — ? or F1 — lists every editor key while the panel is open). Saving never ends the history: undo goes back past the last save, and what you went back to is saved as a new revision, the earlier one staying in History. Inside a text you are editing in place, the same keys walk that text's own history — a burst of typing, a line break, each formatting — until you leave the field, and undoing puts the caret back where the change was made. When a collaborator or the assistant changes the deck while you edit, undo takes back your changes only, never theirs. The history starts over when you leave the deck.

In the panel:

Slides — a filmstrip of the deck: click a slide to go to it, reorder, duplicate or delete slides, change the composition of a slide with its layout picker (the layouts the deck's template offers, every layout for a deck without one; the slide's words move into the new shape and a toast says what was kept, derived or could not find a home), and Add slide with any layout after the current one — the stage lands on the new slide. Add slide and Duplicate stop at the deck's slide limit — 100 slides, or what the deck's template allows (see Templates) — and say so. Everything else about a slide — its look included — is one click away in Selection.
Selection — the slide or the block you selected, with every one of its settings: the title and subtitle, the alignment and the spread of a slide, its theme, its hero and overlay, its speaker notes; for a block, its content (the items of a list, the rows of a table, the tiles of a metrics grid — add, reorder, remove) and its style (size, tone, density, columns, alignment, icons, variants...). A setting left on Default follows the deck's design, exactly as when the assistant writes the slide. A setting that only acts on another shape of the slide — a column ratio on a slide with a single flow of blocks, a spread with one block — says so under its control rather than changing nothing silently; the value is kept and takes effect when the slide takes that shape. Insert a block from the palette here too — text, heading, quote, list, callout, metrics, table, badges, cards, timeline, columns, comparison... Blocks that need a file, numbers or a graph (an image, a logo grid, a chart, a diagram) are the assistant's: ask it in the studio.
Deck — the title, the template and the theme (Dark / Light, among the ones the template supports — see Template and theme), the header and footer pattern (the event, the classification, the footer text, the logo, the footer and the slide numbers — see Header and footer) and the slide transition.

To add a block, click the + under the last block of the slide (or of a column, or of the callout, comparison or columns block you selected) and pick one: it lands at the end of that run, selected. A slide's Selection tab offers the same under Add a block. On a slide with columns, or in a columns block, Equal row heights lines the blocks up as rows: the cards at the same place in each column share one height.

Every change saves itself. There is no Save button: a moment after you stop, the change is written, and during a long stretch of typing it is written every few seconds as you go. The footer of the panel says where your work stands — Saving..., All changes saved (hover it for the time and the revision), Offline (your changes are kept and saved as soon as you are back online), Not saved yet - retrying with Retry now, or Your last change could not be saved with the reason and Undo it when the deck refuses a change (an emptied required field, say). Hiding the tab or leaving the deck saves at once; closing the panel waits for the last save, and asks only if it failed. Your session of edits is one entry in the deck's History, named after what it touched (Edited slides 3, 7, Inserted slide 4, Moved slide 5 to 1); a pause of a few minutes, half an hour of editing, a template change or an edit by the assistant starts the next one. The slides keep the platform's layouts — you arrange blocks, you do not drag boxes around — which is what keeps a hand-edited deck one the assistant can still read, fit and export.

Decks from earlier versions

A deck built with an earlier version of the platform may use slide compositions that no longer exist. It keeps working: such a slide is shown as the closest current composition, with its words, notes and images intact, and is stored that way the next time the deck is saved. Nothing to do on your side — and if you ask the assistant to change one of these slides, its reply mentions the conversion.

Anything that needs judgement or material — a new argument, a chart from your numbers, an image, a whole section — is the assistant's job, and the place for it is the studio; the two work on the same deck, so a hand edit and a prompted edit follow each other in the same History.

Edit in the studio

Studio, in the viewer toolbar, opens a chat beside the deck: the conversation on the left, the presentation on the right as a live preview. Edit with my assistant, on a card in the library, lands on the same screen.

The assistant you talk to already knows which presentation is open — its title, its slides, the revision you are looking at, and what you are allowed to do with it. There is nothing to load and no identifier to paste: "make the title shorter", "add a slide comparing the two options after this one", "warm up the closing" is enough. It also knows which slide is on your screen, so "this slide", "here" and "it" mean the one you are looking at.
Pick any of your chat agents from the menu at the top of the rail, or stay with the General Assistant. Whichever you pick has the presentation tools while it is in the studio, even if it does not carry them elsewhere — you never have to find "the agent that can edit slides" first.
Every change is applied to this presentation, in place: the deck on the right refreshes the moment the assistant saves, the link you shared stays the same, and the change lands in the deck's History like any other edit. The assistant only creates a new presentation if you ask for a copy.
The assistant shows you what it changed. When it edits a slide, the deck on the right moves to that slide as the save lands (unless you are presenting), its reply links the slides it touched — "done, see [slide 4]" — and clicking such a link moves the deck rather than opening a new tab (hold Ctrl or Cmd to open the viewer instead). The saved-receipt line in the rail reads Saved - slide 4 changed. with a Go to slide 4 shortcut. The same links work in the main chat: the inline preview opens on that slide, and a link pasted or shared elsewhere (/presentations/<id>#/<slide>) opens the viewer on it.
The rail is a full conversation: attach a file (a brief, numbers, a logo) with the paperclip, send a follow-up while the assistant is still working to steer it, and scroll up for earlier turns. If the deck came out of a conversation, the empty rail offers to continue that conversation, so nothing the assistant learned while building it is lost.
Make the rail as wide as you read. Drag the divider between the chat and the deck to widen the rail — up to half the screen, for the outline of a whole deck or a slide-by-slide review that would otherwise wrap every few words — and back again; the deck rescales as you drag. Double-click the divider to return to the default width. The divider works from the keyboard too (focus it, then Left / Right, Shift for bigger steps, Home / End for the narrowest and the widest). The width you pick is remembered in your browser for every presentation; it is not part of the link you share.
Collapse the rail with the chevron to give the deck the whole width; a thin Studio strip stays at the left edge to bring it back, with the conversation exactly where you left it. Present leaves the studio first — a chat beside a fullscreen deck would be invisible. While the rail is open, colleagues who have the deck open see you as editing.
The studio is for the people who can edit the deck. Someone who can only view it does not see Studio (nor the strip at the left edge, nor Open in studio on the card in a chat); Copy and edit with my assistant, on the library card, makes them a copy of their own and opens the studio on it.
The studio remembers the conversation in the page address (?studio=1&c=…), so reloading, bookmarking or pasting the link to yourself reopens the same chat beside the same deck.

The keyboard follows the focus: arrows typed in the composer never turn the page, and the deck's shortcuts work again as soon as you click back on it.

Two people, one deck

Several people (and their assistants) can edit a deck at the same time. A change a colleague saves reaches your screen within a few seconds and is merged with yours: what they changed lands, what you changed stays — even on the same slide, as long as it is not the same block. The slides they just changed flash in the filmstrip, and the text you are typing in stays open unless they changed that very block. If you both changed the same thing at the same moment, your version is kept and you are told on which slide. A slide you delete while a colleague edits it comes back with their changes, and you are told: an edit is work, a deletion is quick to redo. While you edit, each colleague's selected block is outlined on your slide in their colour, with their name, and their initials show on the slide they are on in the filmstrip, so you can see where the others work before you touch the same block. The studio follows the same rule from the other side: the assistant reads the current revision before it edits, so a collaborator's change made a minute ago is not silently rewritten.

A deck starts private to you. Share (on the card in the chat, in the viewer toolbar, or on the library card) opens the sharing dialog. Access is granted in the dialog, never by the deck's address on its own (the one exception is its public link, which you turn on explicitly):

share with specific people or groups, choosing Can view or Can edit for each
administrators can publish the deck to the whole company (optionally restricting it to the people and groups listed)
then copy the link to paste into a message

Under the link, Who can open this link says in words who it will open for — only you, the people and groups listed, or everyone in the company — and, while you have unsaved changes, that the new audience applies when you save. Anyone else who follows the link is told the presentation is not available.

Ask the assistant to share it. "Share the Q3 review deck read-only with the EMT: ada@example.com, ..." works in the chat too. The assistant names the people and groups who will get (or lose) access, then shares the deck as you, with the same rules as the dialog: only the deck's owner, or someone it is shared with as an editor, can share it. People are named by the email of their account on the platform (or their user id), groups by name. If an address matches no account, nothing is shared and the assistant tells you which one: it never invites anyone. It can also change someone's access, remove it (they lose access at once) or replace the whole list. Ask "who has access to this deck?" and it reads the current list. Publishing a deck to the whole company stays in the dialog, for administrators. Every change the assistant makes is in the audit log.

Someone who can edit can open the deck, use Edit slides, and ask their own assistant to change it in place from the studio — the link stays the same for everyone. Someone who can view can open, present, export, duplicate and comment on it, and browse its History; the editing tools (Edit slides, the studio, the Layout check) are not offered to them, and their Copy and edit with my assistant action works on a copy of their own, leaving the original untouched. Asked in a chat to change a deck they can only view, the assistant does the same: it offers to copy the deck exactly and change the copy, rather than writing a new deck from what it read.

Sharing never changes the slides: opening the dialog while someone's assistant is mid-edit cannot disturb their work.

Public link

To show a deck to people who have no account on the platform - a customer, a partner, a board, a conference audience - turn on its public link at the bottom of the sharing dialog. The deck gets a second, unique address that anyone who has it can open, without signing in:

it opens the deck read-only: the slides, navigation, the overview and fullscreen presenting, and a link to one slide (#/3) lands on it. A page document reads as a document;
it shows only the slides: no speaker notes, comments, history, sharing or editing, and nothing about who made the deck. Pictures and logos are served only when the deck shows them;
it stays live: an edit to the deck shows the next time the page is opened.

The switch applies at once - it is not part of Save. While the link is on, the dialog shows it with a copy button, who turned it on and when. Replace link issues a new address and the old one stops working immediately (use it if a link reached the wrong people); turning the switch off stops the link for everyone. The people and groups the deck is shared with keep their access either way. A deck with a public link carries a globe on its library card, for everyone who can see it.

Allow PDF download. Under an active link, this switch (off on every new link) adds a Download PDF button to the public page. Visitors get the vector PDF of exactly what the link shows - the slides, never the speaker notes - rendered by the platform when they ask, and reused for a few minutes while the deck does not change. Like the link's own switch, it applies at once. Replace link keeps the option; turning the link off forgets it, so a link turned on again starts without a download. Downloads are bounded per visitor and per link: when too many arrive at once, visitors are asked to try again in a few minutes.

Who can turn it on: the deck's owner, someone it is shared with as an editor (Can edit), and administrators - the same people who can share it. Someone who can only view the deck sees that it is public, but cannot copy, replace or turn off the link.

Ask the assistant. "Give me a public link to the Q3 deck I can send to the customer" works in the chat: the assistant turns the link on (or hands back the one the deck has) and gives you the address, telling you that anyone with it can view the deck. It can also replace the link or turn it off ("turn off the public link of the Q3 deck"), let visitors download the PDF or stop them ("let people with the link download it as a PDF"), and who has access to this deck? says when a public link is on and whether it offers the PDF. It acts as you, with the dialog's rules: only on a deck you can edit, never on a company-published one (that stays with administrators, in the dialog). Every change, by hand or by the assistant, is in the audit log.

Anyone with the link

A public link is a capability: whoever has the address can open the deck, and can pass it on. Turn it off, or replace it, when it is no longer needed. Administrators can turn public links off for the whole platform in Settings > Security > Security options (Public presentation links, or PRESENTATION_PUBLIC_LINKS_ENABLED=false): every public link then stops working at once and none can be created; turning the option back on restores the links that were not turned off.

Work on a deck together

A shared deck is reviewed where it is presented. Three things in the viewer's toolbar are there for the people working on it with you:

Who is here. The initials beside Present are the people looking at the deck right now — hover one for the name, and for whether they are editing and which slide they are on. They appear as someone opens the deck and disappear within a minute of them leaving. While someone edits, the room is refreshed every few seconds, so their changes and where they work show almost at once (see Two people, one deck).

Comments. Comments opens a side panel of review notes. A note is attached to the slide you are on (or to the whole deck, if you say so), so a reviewer can write "this number is last quarter's" on the slide it concerns; clicking a note takes the deck to that slide. Anyone the deck is shared with can read and add notes and reply to them — reviewing is exactly what read-only access is for; you can edit and delete your own, and whoever manages the deck can tidy up any of them. When a note has been dealt with, Resolve moves it out of the way (the author or a manager decides) — Open and Resolved at the top of the panel switch between the two, and This slide narrows the list to the current slide. The count on the Comments button (and on the card in the library) is the number still open.

Your assistant works the comments with you. In the deck's conversation, ask it to "address the review comments" and it reads each open note with the slide it is about and changes the deck in place; ask it to "implement Ada's comment on slide 5 and close it" and it makes the change, then resolves that note — with a reply saying what was done, if you like. It can also leave notes for you ("comment on every slide where a number has no source"), read the resolved ones back, or reopen a thread that was closed too early. Everything it posts appears under your name, in the panel, the moment it is saved, and it can only close what you could close yourself: on a deck you merely view, it will reply to a note rather than resolve it. Nothing is resolved unless you ask — a reviewer asked a question, and you decide when it is answered. A note on a slide that was later deleted stays in the panel, marked Slide removed, so nothing a reviewer wrote is lost.

Writing a note yourself, Generate with AI under the text box drafts it for you: say what the note should convey ("ask for the source of the revenue figure") and the assistant writes it against the slide you are on — it sees the slide's content, not just your words — or against the thread you are replying to. If you have already typed a draft, the same button becomes Improve with AI and tightens what you wrote. The draft lands in the text box for you to edit and post, or discard; nothing is sent until you press Comment.

The panel is as wide as you read: drag its left edge to widen it — up to a third of the screen — and back again, or double-click the edge to return to the default width; the divider works from the keyboard too (focus it, then Left / Right, Home / End). The width you pick is remembered in your browser for the editor, history and comments panels alike.

History. Every change — an assistant generation, a session of edits in the panel, a restore — is kept as a numbered revision, with who made it, when, and what changed (Edited slides, title), the last thirty per deck. Edits by hand save themselves every few seconds, so one person's run of edits is one entry, kept up to date as they go (the revision numbers skip what it folded in). History opens the list; clicking a revision previews it in place, with a banner naming what you are looking at and Back to current beside it. Restore brings a revision back as the current version, and is itself a new revision, so a restore can be undone the same way. Restoring, like editing, needs edit access.

Working alone, none of this gets in your way: the buttons are quiet until someone leaves a note or a second person opens the deck.

Export a deck

Export is available on the chat card and in the viewer toolbar. Every export is rendered by the platform itself from the same slides you see, so the result does not depend on your browser or zoom level:

Format	What you get
`PDF (vector)`	One page per slide, with selectable text and the deck's own fonts. Page size matches PowerPoint's slide size, so a PDF page and a PowerPoint slide of the same deck have identical dimensions.
`PowerPoint (pixel-perfect)`	One full-slide image per slide, with the speaker notes attached and the deck's slide transition. Opens anywhere and looks exactly like the screen.
`PowerPoint (editable shapes)`	Text boxes, cards, tables, badges, buttons and dividers as real PowerPoint shapes placed where the platform placed them — tinted columns, check marks, outlined display numbers, list bullets, callout bars and the rule under each slide title included — so a colleague can keep editing in PowerPoint. Charts are real PowerPoint charts: a bar, stacked bar, line, area, pie, donut, waterfall or butterfly chart opens with its series, categories, colours and number formats, ready to be edited or re-pointed at new figures (see below). Org charts are native shapes: each seat a group of its card or circle (an oval of its size), its tile, badge and words as editable text, every reporting line a connector, the whole chart one group. Icons, diagrams (architecture diagrams included), logos and gradient surfaces are embedded as images — the light along the top of a card and the glow under its figure as a transparent picture under the card's editable text. The speaker notes and the deck's slide transition come along too.

Print to PDF (browser), in the More menu of the card and of the viewer, is the fallback that needs nothing from the server: it opens the deck in your browser's print layout and starts the print dialog. In that dialog pick Save as PDF as the destination: the browser then uses the deck's own page size and every slide lands on a landscape page. Sending it to a printer instead - including a virtual one such as Microsoft Print to PDF - uses that printer's paper (usually portrait Letter or A4), so the slides come out rotated and boxed in the middle of each page.

The print dialog opens once every chart, diagram, icon and logo on the deck is drawn. A figure that cannot be drawn - its code could not be loaded, or it never finished loading - does not hold it forever: after a short wait the deck prints as it stands, and a notice (Part of the deck was not finished) says so when the dialog closes. Use the PDF (vector) export for the complete deck: it never exports a figure that could not be drawn, and fails with the reason instead, so it can be tried again.

Which PowerPoint export?

Use pixel-perfect when the deck is finished and only has to be presented from PowerPoint. Use editable shapes when someone will keep working on it there.

Both play the deck's slide transition with PowerPoint's closest one: Fade fades, Slide pushes the next slide in from the right, Zoom zooms in. The 3D Convex and Concave have no PowerPoint twin, so they become a push and a cover from the right; None cuts from slide to slide.

Charts in the exports. A chart on a slide is data the platform draws, not a picture, and each export treats it that way. The PDF keeps every chart as vector graphics in the deck's own colours and fonts — as sharp as the text next to it. The editable PowerPoint rebuilds bar, stacked bar, line, area, pie, donut and combined bar-and-line charts as native PowerPoint charts: same series and categories, the palette of the deck's template, the same axis range and the same number formats (€16.5M, 86%, 12,400 users), the legend and the target line where the slide had them — and a chart that plots two units keeps its right-hand axis as a second PowerPoint value axis. The values written on the bars are PowerPoint's own data labels, at the size the slide set them, turned on their side where it turned them and a row higher where it wrote them on two rows - further along their row on a horizontal chart - the bars spaced as on the slide; a line's value the slide moved clear of a bar, another value or the axis sits where the slide wrote it, and a line's values written alternately above and below the points are PowerPoint's own labels above and below them; on a stacked bar, the segments that carry a value on the slide carry it in PowerPoint, and the stack totals sit above the stacks as text - past the end of the bars on a horizontal chart, inside the chart's frame as on the slide. A waterfall is an editable stacked bar chart in which each bar stands on an invisible base series — the way a waterfall is built by hand in PowerPoint, and one that opens in Keynote, Google Slides and LibreOffice too — with the slide's colours, its signed values and its connector lines; to change a step in PowerPoint, edit its bar and the base under the steps after it. A butterfly is an editable clustered bar chart: the left-hand group is stored as negative numbers so its bars run to the left, and a number format hides the minus sign, so labels and the axis read plain sizes and a 0 has no label - to change a value on the left in PowerPoint's data sheet, type it as a negative number. PowerPoint's bar chart has no column between its two sides, so a butterfly whose level names sit on the spine on the slide shows them on the left axis in PowerPoint, and the export says so in its notes. Three figures have no PowerPoint equivalent — a gauge, a map, and bars read against two different axes (PowerPoint would draw them over each other) — and are placed as a high-resolution picture of the rendered figure instead, a map's title, caption and legend staying editable text beside it. When a stand-in is used the export says so: a short note appears when the download starts (Exported with notes, naming the slide and the chart), and an assistant that exported the file for you lists the same notes in its reply, so nobody opens the file expecting PowerPoint's own waterfall chart, or to edit a chart that is a picture.

Move a deck to another platform

The last entry of the Export menu, JSON package (import elsewhere), is not a rendering: it is the deck itself — its slides, the name of its template and every logo, diagram or image it uses, embedded in one <title>.presentation.json file. Import on the Presentations page reads that file back and creates the deck under your name, so a deck built on a staging or local platform continues on production, and a deck can be kept as a backup outside the platform.

On import, the platform first looks for what it already has: a template with the same name (a built-in brand deck the destination also ships, or a company template you recreated) and, inside it, the assets by filename — so a deck that uses the brand logos points at the destination's own logos and nothing is duplicated. When the template holds several files of that name (a dark and a light logo.svg), the one with the same content is used; if none has it, the one the template lists first is used and the confirmation says so. Only what is missing is copied out of the file, as files that belong to you. The confirmation says what happened: how many assets were matched, how many were copied, and which template the deck ended up on. If the template does not exist on the destination, the install's default template is used and the message says so; if a file can be neither matched nor copied, the import is refused naming it, rather than silently producing a deck with a hole in it. A picture that was already missing from the deck when it was exported is the exception: the deck imports without it, and the confirmation says how many.

The imported deck is a new deck — revision 1 in its history, with none of the original's shares or comments — and validates like any other: a package written by a newer platform whose slides use a layout the destination does not know yet is refused with the exact slide and field. A package embeds only the files you could download when exporting; a shared deck exports the files its owner shared with the deck, never anything else of theirs.

Template and theme

Two words describe how a deck looks, and each means exactly one thing:

The template is the deck's design: a palette per theme, the heading and body fonts, the logos drawn on every slide, the predefined assets the assistant may place, and a few sentences of composition guidance. A deck wears one template, or none.
The theme is the colour scheme the deck is drawn in: Dark or Light, nothing else. A template supports the themes it has a palette for — a template designed only on a dark background supports Dark alone — so the theme picker of a deck only ever offers what its template supports. A deck with no template offers both.

Out of the box every platform ships with the read-only XTM One Brand Deck (the product's purple palette, Dark theme only). Creating your own — by hand, or from a brief and a logo with Generate with AI — is covered in Templates.

Choose a template when you create a deck

New presentation, on the Presentations page (and New presentation with this template on a template's page), asks which template the deck starts in — one of the templates you can see, No template - plain look, or Let my assistant ask me in the chat — the theme among the ones that template supports, and which assistant writes it (Written by: only the assistants that carry the Presentation Generation tool are offered, so the chat you land in can always produce the deck). It then opens the chat with the request written for you; add what the deck is about and send.
In a chat, when you ask for a presentation and you can see several templates, the template is your choice, never the assistant's: unless your request names one ("in the Sales template") or asks for "no template" (the plain built-in look), the assistant lists your templates and asks which one to use before creating anything — the platform refuses to create a new deck in a template you did not pick, even one the assistant would have chosen for you. It may recommend one, but the deck is created once you answer. With a single template in view, it is used and the reply says so.
Automations, and chats where the assistant cannot ask, use the default template an administrator sets in Settings > Parameters — the platform's built-in brand deck unless another company template was chosen.
On XTM One, the CTEM Assistant is the General Assistant of the platform and carries the presentation tools out of the box, as do the XTM-One Assistant and the assistant of each product (OpenCTI Assistant, OpenAEV Assistant, OpenCRQ Assistant, FIMI Analyst), with the Presentation Design skill; on the Copilot edition the General Assistant does, and any agent an administrator gives the Presentation Generation tool to is offered as a writer.

Change the template or the theme of a deck

A template is not copied into the deck: the deck stays linked to it. Its colours, fonts and logos are read from the template every time the deck is drawn, so editing the template restyles every deck wearing it, at once. Everything else — the slides, their words, the images you uploaded, the theme you picked — belongs to the deck.

Edit slides, in the viewer, opens the Design section: pick another template from the Template menu (or No template - plain look) and the deck restyles on screen at once, then saves itself like any other edit; the Theme toggle beside it offers Dark and Light and greys out what the chosen template does not support — switching to a template with a single palette snaps the deck to that theme. In a chat, "apply the Sales template" or "switch this deck to the light theme" does the same; the assistant tells you when a template cannot draw the theme you asked for. Deleting a template does not touch the decks that wore it: they keep their content and fall back to the plain look.

A slide with its own theme

A slide can be drawn in the other scheme than the deck: a dark cover and a dark closing slide over a light body is the classic shape of a brand deck. In the Selection tab of Edit slides, Theme offers Inherit, Dark and Light for the slide. Inherit — the default — follows the deck theme; Dark or Light fixes the slide's own scheme, and everything on it follows: its paper, its text, its tones and charts, and the logo variant drawn in its corner. A slide with a theme of its own keeps it whatever happens to the deck theme: switching the deck between Dark and Light changes only the slides that inherit (the Design section says how many slides keep their own theme), and the slide comes back to the deck only when you choose Inherit again. The template's rule applies to slides too — a slide can only take a theme the template supports, and a template that cannot draw a slide's theme snaps that slide the way it snaps the deck. In a chat, "make the cover dark" or "let slide 5 follow the deck again" does the same; the assistant's outline marks the slides that keep their own theme.

Covers: hero backdrops

A title or a closing slide has a hero — the backdrop behind its headline, drawn in the slide's own colours: Plain (the flat paper), Gradient (the brand gradient washed over the paper, with the template's mark as a faint watermark in the corner — the look of a brand cover), Split (the text on the left, a gradient panel on the right that carries the slide's background image when it has one), Spotlight (a soft glow behind the headline and two thin arcs in the corner). A template can add covers of its own: the THREAD Conference Deck adds the tiles of the THREAD design kit — Mosaic (coloured square tiles gathered in the corners, clear of the headline, the logo and the slide number — each slide has its own arrangement, the same every time it is shown or exported) and three more shapes of the same tiles: Cascade (tiles stepping down from the top-right corner, the logo large in the top-left corner and the headline low on the slide, over a thin rule and one row with the speakers on the left and the footer line on the right), Horizon (a skyline of tiles along the bottom edge, the logo large in the top-left corner) and Pillar (a column of tiles down the right edge). These covers are THREAD's artwork, so they are offered on THREAD decks only: a deck on the Filigran Brand Deck, the XTM One Brand Deck or no template never shows them, and a cover that carried one (a deck moved to another template, or made before this rule) is drawn on its template's default cover — Gradient on both brand decks. Pick the hero in the Selection tab; a dark gradient cover on a light deck is the slide's theme set to Dark plus the Gradient hero. Under the hero, Hero colour says which colour of the template's palette the backdrop is drawn in — each chip is painted in the exact colour the slide will use: Brand gradient (the default — the template's gradient, from its first colour to its second), or one colour alone: Accent, Gradient start, Gradient end, one of the Info / Success / Warning / Error tones, or Muted. On the XTM One Brand Deck the gradient runs from the product purple into cyan, so a cover that should stay purple picks Accent (or Gradient start); an incident review can open on an Error red spotlight. On a title slide, Accent the last line of the title colours the last line of the headline (after a line break — or its last word) in the accent tone, so "Uncover Threats. / Take Action." reads the way the brand deck does without any formatting in the text. Any slide with a background image can also take an overlay — None, Soft or Strong — a scrim between the picture and the text so the headline stays readable. Templates can set the cover defaults for new decks — the hero, its colour and the slide's theme (the Filigran Brand Deck and the XTM One Brand Deck open and close on a gradient cover) — and everything exports as it shows: the PDF draws each slide in its own colours, and the PowerPoint carries the backdrop as a picture under editable text.

Speakers and icons on the title slide

A title slide names up to four speakers — a keynote's one, a talk two people give together, a panel — side by side under the headline, in speaking order, each with a name, a title, up to three credentials and an optional portrait. Ask for them ("Samuel Hassine and Julien Richard present, with their photos") or edit them in the Selection tab: Speakers adds, reorders and removes them, and a speaker's Photo is picked from the people photos of the deck's template. A deck made before several speakers existed keeps its one speaker, now the first of the list. The closing slide still names one speaker, with contacts.

The row of icons under a title is edited in the same tab: Icons lists them in order — pick another icon by typing its name (suggestions appear as you type), reorder them, remove one, or add up to six.

Header and footer

The line at the bottom of every slide is not written on the slides. It is drawn from a handful of deck settings, so it is set once and never edited slide by slide — and it cannot drift: a footer that reads the deck's title changes on every slide the moment the deck is renamed.

A template can define the footer (and, optionally, a header) for every deck wearing it, as a pattern such as {{event}} - {{title}} - {{classification}}. {{title}} is the deck's title, {{subtitle}} the subtitle of its opening slide (as plain text — the footer never shows inline emphasis), {{event}} the event you set; anything else — a classification, a client, a version — is a field the template declares with a label and, optionally, a default. Both built-in brand decks define {{event}} - {{title}} - {{classification}}; like the event, the classification is empty until you set one. An empty part disappears together with its separator, so a deck with no event reads Quarterly review - Confidential, not - Quarterly review - Confidential, and a deck with neither reads Quarterly review.

Edit slides, in the viewer, has a Header & footer section:

Show logo, Show footer (and the header, when the template defines one) and Show slide numbers switch each part of the furniture on or off for the whole deck. Opening, section and closing slides never carry the footer; a body slide can leave it out on its own (hide_footer).
Event names the occasion — a conference, a customer meeting, a board date.
One field per variable the template declares (Classification for the brand deck), labelled by the template, with its default shown greyed until you type. A deck wearing a template with no pattern, or no template at all, gets a free Footer text instead, drawn after the event.
The Footer preview line shows exactly what every slide will read as you type; the panel tells you when the footer switch hides it.

In a chat, "mark this deck Confidential", "set the event to FIC 2026" or "hide the slide numbers" changes those settings — one field, not thirty slides — and the assistant's reply says what the footer now reads. Exports (PDF and PowerPoint) draw the same line. How a template declares its pattern is covered in Templates.

The Presentations page

Work > Presentations in the sidebar lists every deck you can see, most recently changed first and a page at a time; the page bar sits above and below the grid, like on the Videos page. Search by the words of a title, the owner's name or the template's name, show only Decks or Pages, and narrow the list to Mine, Shared or Company: the search and the filters cover your whole library, not only the page shown, and the count next to them is how many decks they match. The page, the filters and the search stay in the address, so a link or a reload keeps them. Each card shows the slide count, the template it wears and its theme (Dark or Light), the owner, when it last changed, and - as icons on the right of the attribute row - whether it is shared with you, company-managed, or view only. Click anywhere on a card to open the deck. The card's footer carries Present (or Open for a page), Edit with my assistant (which opens the deck with the studio beside it), and a menu with Share, Duplicate, and Delete for a deck you manage. Import, next to New presentation, reads a JSON package exported from another platform (see Move a deck to another platform).

Deleting a deck removes it for everyone it was shared with; a chat message that embedded it then shows the deck as no longer available.

Next step

Set up the brand your decks are generated in with Templates.