Guides

AI Home (the assistant), ContentCanvas (your library and working board, the home screen), Tasks, Automations, Brand, Integrations (connected apps), then Train and Produce side by side, one flow read in that order. Below them sit search, Settings, this Help drawer, and your Account.

Press Cmd+K (Ctrl+K on Windows) at any time to jump to any destination, any Job, or any piece of content by name, from anywhere in the app.

Start a production run from a source on the ContentCanvas board, produce a single asset from the Produce page, or hand the work to AI Home in plain language. All three end in the same place: real, editable content filed into ContentCanvas with a record of what it was built from.

Drag a file from your machine straight onto the ContentCanvas board. It uploads, shows itself being read, and lands as a source split into sections. PDFs, documents, decks, images, audio, and video are all accepted; a file the platform cannot take says so plainly rather than failing quietly.

Point the platform at a URL to read one live page, import a whole website's content in bulk, or run a research job to have it gather and cite current material across the open web. All three arrive as ordinary sources in the same library.

Connect a third-party app on Integrations, on the left rail. Its items then appear as their own tab in ContentCanvas, and importing one turns it into a source split into sections, exactly like an upload.

The tabs across the top are grouped by where things came from: what you singled out (Starred, Folders), what you brought in (Source content, Research, Website), what is plugged in (each connected app), and what the platform made (Generated, Collections). Each card carries a cover, its format icon, and its stage.

Switch layouts with the toggle: Canvas is the spatial board and the default, Grid shows covers, List is dense for scanning, and Graph shows lineage so you can see which assets came from which source. On a phone you land on List.

Type in the search box to find anything by keyword or by meaning. Filter Generated work by workflow stage, or open Collections to slice the whole library by the kind of material inside it. Clicking a source on the board fans out everything made from it; clicking a production run opens what that run produced.

Tabs, layouts, the thing you drilled into, even the section you selected are all in the address bar. Back and Forward work, a refresh restores exactly what was on screen, and you can send a colleague a link to precisely what you are looking at.

Open Train and answer one question: what does this design make? That name (a Webinar promo, an Event one-pager) becomes the template's name and the name of everything it produces. Then drop in the finished piece itself: an image, a PDF of up to 20 pages (a multi-page PDF trains as a deck, one slide per page), an HTML or Word document, or plain text. Press Train this template.

Training is a tracked run you can watch in your Activity panel, and it takes a couple of minutes. The platform reads the design, lifts every word, logo, and button off the artwork while keeping the artwork itself, names each text element the way a designer would (Headline, Subhead, Call to action), and rebuilds the whole thing as an editable template. You can close the tab: the tile on Train shows live progress and survives a reload.

Every template is a tile: the cleaned artwork as its cover, its name, and how many fields it has to fill. A tile still building says so, and a failed one tells you why in plain words. Open a ready template to fine-tune it in its editor: move and resize elements, restack layers, adjust type and color. The base artwork is the frame's background, so you can never grab or delete it by accident.

The moment a template is ready, a Make Job named after it appears for your whole team: in production runs, in AI Home, and in every Automations picker. Removing a template stops it being offered, nothing more: the style it learned and every asset it produced are kept.

Open Produce and pick the template, then the source it should be about, and optionally the exact material inside that source. Draft the copy and every field fills in, ready to edit, with the original copy shown beneath each one: the original is the specification for how much text that space holds. Nothing is made until you press Proceed, and then you land straight in the finished asset's editor.

An image field offers three ways in: pick from your existing content and brand assets, upload a file, or generate with AI. On a graphic, a field you leave empty leaves that element off the asset rather than rendering a blank; an HTML or text template keeps the picture it was trained with.

On a graphic template you can change the artwork itself before producing. Drag a box over the part of the picture you mean, describe the change, add as many regions as you like, then apply them in one AI pass. The edits are confined to the regions you drew; the rest of the artwork is untouched.

To produce a batch, start a production run from any source: from its card on the board, from its page, or from AI Home. The platform pulls the source's usable material, you keep all of it or pick the parts that matter, then choose one or more templates; each becomes one asset, built as a live tile under the source. The run belongs to the server, so you can close the tab. If a template you picked is still training, the run waits for it and shows the training's own progress in the tile.

Every produced asset carries a small plus beneath it on the board: describe the changes and a new version is produced, chained under its parent. And an uploaded image of a finished design can become an editable copy in one click with Create a new version of this source; the platform trains the template the first time and reuses it after that.

The asset opens in View: a clean proof of the finished work on a dark mat, at the size and shape it really is. Toggle to Edit in place. There is no separate viewer and editor to keep straight, and the preview adapts to the medium, so a deck pages as a deck, a document reads as a document, and video and audio play.

Covers are made from the asset itself, not a screenshot of a page. A LinkedIn post covers as the feed card it will be: your brand name and avatar, the copy folded where LinkedIn folds it, the creative edge to edge, and the reaction bar.

Each asset records what it was built from, so you can trace any piece of finished work back to the source, the run, and the template that made it, from the asset page or from the Graph layout.

Download the asset in the formats its type supports, take everything a run produced with Download all, or mint a share link that anyone can open with no account. The platform does not post to your channels on your behalf: you take the finished file and publish it yourself.

Most deliverables open into the section editor: the outline on one side, a live preview on the other, kept in step. Click something in the preview to find it in the outline and the other way round. Click a heading or paragraph and type; there is no Edit button and no overlay.

Ask AI to rewrite, shorten, or restyle one section without touching the rest. Every section keeps its own version history, so you can compare what changed and restore an earlier version.

A designed graphic opens on a spatial canvas: drag elements, resize with the handles, restack the layer list by drag (it shows the real front-to-back order), add pages, apply a style, or reflow with the layout picker. A deck pages slide by slide, each slide fitted to the pane. An image opens to crop, rotate, replace, regenerate, or refine a selected region with AI, archiving each version as it goes.

A video opens on a timeline to trim a range, grab a frame at the playhead as a new image, or export a section. Voiceovers and music open in the audio editor.

The Brand Voice tab holds your positioning, your tone, and the phrases that must always or never appear. Every job that writes anything reads it first.

Brand Assets holds your colors, fonts, logos, and reusable artwork, with bulk upload and an import that reads a PDF of brand guidelines. If the platform has never been given a logo, it will never invent one.

Build one persona per buyer type in your committee, and list the products or services you sell. Attach a persona and an offering to a run and the asset is written for that buyer, about that offering, instead of being generically about your company.

Under Brand, Training, Examples, drop finished files in. You do not have to tell the platform what they are first: it looks at each one and works out what it is, and you only correct what it got wrong. One drop can teach several formats at once, and a logo dropped in is recognized as a logo and routed to your Brand Kit.

Each file is read by four separate passes, and a pass only runs where it has something to say. Open an example to see the findings: the file itself beside the individual elements the platform picked out of it, and what it concluded from each. You can see the evidence, not just the verdict.

Brand, Training, Styles is the Style Studio: everything Training has distilled, read back to you. The specimen at the top is painted in your own captured colors with your own voice quoted in it. Each learned rule is a card carrying its source, so you can see which example taught it, and what changed. Tokens it has never seen show as empty, never as a color you do not have.

Tap a swatch on the specimen to correct that one token. Run a test to see something made to your style right now, side by side with the example that taught it. The coverage strip shows one chip per format, so you can see at a glance what it has learned and what it has never been shown; a rule can be scoped as narrowly as one offering in one format.

An asset that has never been measured shows an invitation to check it, on the asset page and in the editor. The result is scored on four axes (color, type, logo, layout) rather than as a single number, so a low score tells you where the problem is, and it says what it measured: the whole asset, or just the part you last changed.

The check is observational. It will never stop you approving or downloading anything. A score produced by an older version of the check is labeled out of date with an offer to run it again, rather than being presented as current.

Send for review from the asset page, the task list, or the board. You name the people who have to sign off, and the asset waits on each of them until they respond. Reviewers outside your organization get an emailed link and can approve or request changes without an account.

An asset's status is a control you set, not a label you read. Changing it is how work moves from draft to finished. A Changes requested card says what was actually requested, not just that something was.

Every asset keeps its approval history: who was asked, who responded, what they said, and the latest change request pinned to the top. A reviewer whose invitation failed to send is shown as failed with a retry, rather than sitting there as though they are still thinking about it.

Any Make Job can run on a schedule: pick the content it should draw from, optionally give it an angle, and each fire produces one finished asset in your trained design. Run it once at a set time or on a cadence; the calendar is the default view, and any finished run can be turned into a schedule with Repeat.

On a list works down a queue, one item per run. On trigger fires when something happens, and the useful one is a new source landing: when new content arrives, make my trained design from it. Both file their output into ContentCanvas like any other run: a schedule produces assets, it does not post them anywhere.

The Ideas view is your planned backlog, where content ideas you kept from an ideation job wait to be produced. History keeps a record of every execution, so you can see what ran, when, and what it made.

Brand, Integrations, Train, and Produce are rail destinations, not Settings tabs, so most members rarely need Settings at all. Organization admins manage Users, Models, and Billing there, and some workspaces also carry Usage and Performance.

This Help drawer holds a live catalog of every Job your workspace can run, our privacy and security documentation, and a contact form. If the platform is down or in maintenance you will see a notice in the app itself, and the system status page is linked from Contact.

The unit of production. On Train you hand the platform one finished design your team is proud of, and it learns the style, the layout, and every fillable field from that file. Your workspace can only produce what it has trained: the platform never invents a design, so everything it makes looks like something your team made.

What you actually run. Training a template automatically mints its Make Job, one marketing outcome with the design already decided. Alongside those, the platform ships Jobs that plan, research, and measure. Pick one, fill in a few details, and run it.

Underneath, a Job is a sequence of single operations: read the source, research the topic, draft the words, fill your template, file the result. You never run those one at a time. The Job runs them for you, in the right order. What it can do, further down, is the full inventory of the operations it has to work with.

Start a production run from a source on the ContentCanvas board, produce a single asset from the Produce page, or hand the work to AI Home. Everything produced is filed straight into ContentCanvas as real, editable content, with a record of what it was built from.

Nothing is generated in the abstract. Your brand rules, your buyer committee, your offerings, and everything Training has learned from your own work are injected into every step, and the finished asset is measured back against them by the brand check.

Compose a polished written document from a source.

Grow a one-paragraph idea into the finished piece.

Render selected clip segments into a finished video.

Recommend the right content to make from a source.

Compare your ContentCanvas library against what you have actually posted on LinkedIn, and get a gap analysis plus a ranked list of repurposes to publish next.

Set a monthly budget and average deal value and get a full-funnel B2B media mix model: AI proposes the channel mix and benchmark assumptions, the platform computes spend → leads → opportunities → sales → revenue → ROAS, and files it as an editable, exportable plan in ContentCanvas.

Point at a real LinkedIn profile and get a sharp, usable B2B buyer persona — their priorities, pains, how to reach them on LinkedIn, and the messaging that lands — grounded in that person’s actual role and background.

Pull latest news and messaging for a competitor

Point this at a target account's LinkedIn company page and get an ABM-ready enrichment: their public shape (size, industry, product motion), what their people are posting about, where the openings are, and 3 to 5 outreach angles tied to observed evidence.

Pull the quotes, stats, and key points from a source.

Turn one piece of source content into five social post ideas, each a hook, a one-line angle, the buyer persona it speaks to, and why it matters.

Get a ready-to-implement paid LinkedIn campaign plan: the recommended objective, a target audience built from real LinkedIn facets and your buyer personas, budget and schedule, the creative to run, tracking to set up, and the exact Campaign Manager steps.

Research any topic into briefed, build-ready notes.

Point this at a competitor's LinkedIn company page and get a briefing on what is working for them: their best-performing post formats, the hooks and angles that earn engagement, where the gaps are, and how to respond.

Point this at your own LinkedIn company page and get a ranked read of your best-performing posts: the hooks and angles earning engagement, the formats that landed, and what to double down on next.

Push a change set to a connected publishing app.

Push a change set to any connected publishing app the org has authorized (Composio).

Analyze metrics and surface what actually matters.

Pull performance metrics from a connected app.

Read a pasted, uploaded, or selected source into clean markdown the rest of a job builds from, keeping a reference back to where each part came from.

Research an objective across the open web and return one grounded, cited briefing. Never invents a claim.

Find the best existing content in your library for a topic, using meaning as well as keywords. Read-only.

Read a public LinkedIn company page and its recent posts, rank them by engagement, and file the result as a source.

Read a public LinkedIn person profile and file it as a source document, ready to build a persona from.

Compare what is in your ContentCanvas library against what you have actually posted on LinkedIn, and report the gaps plus ranked repurposing ideas.

Produce a ready-to-implement LinkedIn ad setup: objective, facet-based audience, budget, creative, tracking, and the Campaign Manager click-path.

Turn a video or audio clip into a transcript with a word-level, timecoded timeline.

Pull performance data from a connected ad account as structured, time-series records.

Mine a source for quotes, stats, testimonials, insights, and headline lines, each grounded in the source.

Given existing content and the buyer persona it targets, propose a ranked set of repurposing ideas, each with a detailed rationale and the Job that produces it.

Propose several content-idea briefs, pause for you to pick the ones worth keeping, and file those to the Planned backlog on Automations.

Turn performance data into a plain-language read plus ranked, actionable recommendations with a confidence on each.

Select the strongest contiguous segments of a transcript to keep, as an ordered edit list.

Inject the Brand Kit (voice, required and banned phrases, colors, fonts, logo) as a directive every AI step follows.

Inject the chosen buyer persona so the output speaks to that buyer's pains and language. Can fan a run out into one asset per persona.

Read one finished design into a reusable template: keep the artwork, lift off every word and mark, name each fillable field the way a designer would, and learn the style and layout. A ready template mints its own Make Job.

Fill a trained template with new words and images drawn from your material, at the template's exact geometry. Every production run ends here, so what comes out looks like what you trained.

The authoring engine for composed work: writes the finished, on-brand artifact a Job asks for, as a flowing document or a designed page.

Compose a titled, sectioned document from a source, grounded in it and steered by a brief.

Restructure a document into your house style without changing a single word. Built for legal, HR, and compliance text.

Turn raw numbers into one structured chart (bar, pie, table, or metric). Never invents figures.

Compute a full-funnel media mix model from a budget and an average deal value, filed as an editable plan.

Rebuild a flat image as a pixel-faithful, editable page with live text layers.

Rebuild a flat image as an editable spatial graphic you can drag and restyle.

Generate promo imagery from a scene description, optionally guided by one or more reference images.

Generate a short video clip from a supplied still image.

Synthesize a spoken-word voiceover from a script, in your organization's voice.

Generate a short, licensing-safe instrumental clip from a prompt.

Render a finished MP4 from an edit list, with an optional branded intro and outro.

Run a plain-language objective across your connected apps, discovering and calling the right tools across as many apps as the task needs.

Would push an approved change set to a connected destination behind a human go-live approval. Not enabled: nothing leaves the platform without you taking it out yourself.

Would publish one filed asset to a connected LinkedIn company page behind an approval. Not enabled.

Would upload an image creative into your LinkedIn Ads asset library. Not enabled.