Skip to main content
A Brand Kit tells Memo’s agents who your brand is. Colors, fonts, reference images, guidelines — encoded once, applied everywhere. Knowledge should live in systems, not people.

​
Browsing Brand Kits

Use List / Grid in the Brand Kits overview. If a search has no matches, choose Clear search to return to the full list. List uses the same rows as Collections, with a thumbnail, kit name, asset count, readable updated date and an arrow to open the kit. The row menu retains Delete when your role permits it. The creation row uses the same layout. Your browser remembers the choice across Brand Kits and the reference asset sections.

​
Where Brand Kits live

Open Brand Kits from the sidebar. This is where every kit, asset, and skill lives across your workspace:
  • Brand Kits — grouped identity, guidance, collections, rules, and skills
  • Cinema Kits — the story, cast, locations, props, camera and voiceover of a film or ad. See Cinema Kits
  • Skill library — reusable creative recipes, including general skills, brand skills and Memo starters. See Skills
Open any kit to see and edit its full detail — assets, colors, fonts, and voice & rules.

​
Creating a Brand Kit

From Brand Kits, use the empty-state Create Brand Kit action (or the New Brand Kit card once the library has content). Give it a name and it’s created empty. A kit page has three views:
  • Overview — the brand at a glance: logo variants, palette, typefaces, and the brand guidance laid out by what each part is: the lead paragraph with where it came from, taglines as quotations, voice, audience and vocabulary as cards, and the visual system (look, color, type, imagery) as a block. Hover any part for its pencil to edit just that part in place; the section’s own pencil rewrites the whole document. Page references link back into the brand book. A completeness meter at the top says exactly what is still missing (“Only 2 colors”, “No font is set for headings”, “Memo has no file for Söhne”) with the one button that fixes it; what is already in place shows as ticked chips under it.
  • Collections — the kit’s images and videos, organized into named groups. Unfiled holds assets saved to the kit that haven’t been added to a collection, and only appears while it has something in it.
  • Guidance & skills — the brand’s Do / Don’t rules, inherited workspace guidance and reusable playbooks in one place. Rules are plain sentences. Only narrower subject or collection conditions and inherited sources need a label. Skills provide procedures for particular tasks.
Open the kit to configure:
  • Logo — the kit’s primary visual identifier, used throughout Memo surfaces
  • Assets — reference images that represent your visual style (logos, product shots, style references, typography samples)
  • Colors — your brand palette, with near-duplicate dedup and neutral filtering built in
  • Fonts — your typography preferences, deduplicated and capped
  • Voice & Rules — a single free-text document describing your brand’s voice, personality, banned language, audience, and hard rules. Markdown-friendly, no fixed structure. Write it the way you’d explain your brand to a new hire.

​
Kit thumbnail

The picture a kit shows in the Brand Kits grid. Set it from the kit page by choosing the image beside the kit’s name, or from the kit card’s ⋯ → Set thumbnail. Until you set one, the card previews the kit’s own assets.

​
Filing assets into a kit

From an asset’s menu inside a kit collection, choose Add to collection. The same picker is available from Inspirations and generated results. From a kit it opens on that kit’s collections: check the ones you want. The arrow beside the kit’s name lists your other kits, and General collections switches to the collections in Inspirations. Each check is saved as you make it, without waiting, and you can file the item in several places. New collection creates one in the list you are viewing and adds the item. Choose Done to close. Logos go to Overview → Logo. Product images, style references, characters, graphic elements, photography, and videos can live in collections. Filing preserves the asset’s type and reuses the existing asset; it does not make a second media copy. In the kit’s Collections overview, use List / Grid to choose landscape previews or compact rows with asset counts and creation dates. This uses the same card family as general Collections and Studio. Newest collections appear first after New collection (and Unfiled, when it has assets). Memo remembers the Brand Kit collection layout independently of Studio, Workflows and general Collections. Inside a collection, Add assets lets you choose from Your library, which lists every image and video in the workspace, including older references that are not in any kit yet, or Upload images. In Unfiled or an existing collection, open the asset’s menu and choose Add to collection to organize it. Open an asset to see its details; videos open with playback controls. Editors can update details, and viewers can inspect them. Remove from collection unlinks the asset. It remains in any other collections, or returns to Unfiled when its last collection membership is removed. Deleting a collection follows the same rule; it does not delete the underlying media. Organizing assets does not automatically attach every collection member to a generation. Videos saved here are available for browsing and playback; saving a video does not make it an image reference for the AI.

​
Voice & Rules

The Voice & Rules editor remains one free-text document per kit for background, voice and creative direction. Put requirements that must be checked for every applicable task into the separate rules in Guidance & skills, so their scope and priority can be inspected. When you @mention a brand kit in Studio, the agent reads your Voice & Rules alongside your colors, fonts, and reference images. Write what matters. Skip what doesn’t. Examples of the kinds of things worth writing here:
  • Voice qualities — “warm but precise, never cute”
  • Banned language — “no em-dashes, no ‘unleash’, no ‘revolutionary’”
  • Audience cues — “we write for founders, not marketers”
  • Visual rules — “always natural lighting, never stock photo aesthetics”
  • Product background — the product range, materials and distinguishing features
A short auto-generated summary is also stored alongside the full text. The summary is what travels into the agent’s ambient context on every generation; the full text loads when you @mention the kit. This keeps short prompts lean and long prompts rich. Every rule belongs to a Brand Kit: add it in the kit’s Guidance & skills, or on one of its collections for a single product line or campaign. A requirement you want across several brands goes into each of those kits.

​
Collection rules

Open any collection inside a Brand Kit and it becomes its own page: a back arrow to the collection grid, the name, a count line, and one Add assets action. A toggle switches between Images and Rules. Collection rules use Do and Don’t columns. Keep each rule focused on one requirement. Click its text to edit in a multiline field, then choose Save rule or Cancel; failed saves retain your text. Long rules expand with Read full rule, and narrower scope labels appear as plain text below the sentence. New rules are required by default, so there is no Required or Brand-wide button to configure. Older preferences retain their meaning and a plain Preference label. The agent receives instructions that required rules outrank preferences and more specific guidance wins among equal priorities. This is a precedence policy, not an automatic semantic conflict detector or a guarantee that an image model obeys every constraint. Collection rules apply when the collection is explicitly selected for a task or its referenced assets establish matching context. Inherited guidance opens the workspace and brand-wide requirements alongside the collection’s own rules. Conditional subject rules are selected from matching references. A rule with both collection and subject conditions can match either condition. Skills live in the workspace or Brand Kit library. The agent chooses a recipe by the task, using its name and use description; there is no skill recommendation control inside collections. Existing skills and references remain saved. Older collection associations no longer affect discovery, and connected agents cannot create new ones.

​
Using rules in Studio

Create with this brand, at the top of the kit, opens a new Studio project with this kit already chosen in the chat. See Starting from a Brand Kit. Editors and owners see the button; viewers do not. In Studio chat, choose your Brand kit and attach the assets you want to work from with @. Brand-wide rules apply with the kit; collection and subject rules follow the assets used for each generation automatically. The composer keeps these choices together without a separate collection mode or context panel. Studio and connected-agent generation use the same rule-selection logic. The final check runs after the actual references are known. Required rules are not silently removed to fit the context budget: if they do not fit, generation is blocked before submission and the requirements must be narrowed or edited. Preferences may be omitted to leave room for required rules. Authoring checks do not spend generation credits. Unfiled never carries rules or skills. It is a holding area, not a product, and it disappears once every asset is in a collection.

​
Brand assets

Upload reference images and Memo handles the optimization — images over 5MB are automatically resized, JPEG/PNG transparency is detected, and everything is stored at the right resolution. Run analysis on any asset to extract its color palette. The extracted colors are deduplicated (RGB distance < 18), capped at 12, and saved to your brand palette. You can set any asset as the default for its type (one default per type per workspace).

​
Characters

A character is a person or character you want to appear consistently across generations. Characters live in kits. Open a collection in a Brand Kit or Cinema Kit and choose Add assets:
  • New character — give it a name, an optional description and one or more images. Memo saves it into the collection you started from.
  • Your library — add a character you already have to this collection as well.
The Character Sheet workflow makes characters too: give one image and get a full reference pack (turnaround, model sheet, expressions, poses, details). Save as character keeps the run as one character, then asks which collection to keep it in. Select a character inside a kit to open it. There you rename it, change its description, add or remove images, or delete it. One character can be in several kits: it is the same character everywhere, so an edit shows in each kit and Delete character removes it from all of them. You can undo a delete for a few seconds. A character with several images is a character reference pack: the first image is the cover, the rest are its sheets. This is one reusable character asset, separate from the folders called Collections.

​
Bring your brand with you (onboarding)

Right after you name your first workspace, onboarding asks how your brand should come in. Four doors, one review — nothing is saved until you confirm it:
  • Scan your website — paste your homepage. Memo reads the site’s own code (not a screenshot): the colors its stylesheets actually use, the typefaces it loads, the logo from the header, the app icon, the social preview and hero imagery — then follows the main menu into the site’s own sections so each becomes a collection, with the copy from every page feeding voice and rules. Takes about 30 seconds and lands in the same review as a brand book. See Scan your website.
  • Upload your brand book — PDF, PNG or JPG, read cover to cover with every printed value verified. See Auto-extract from a brand book.
  • Import from your AI agent — Claude, ChatGPT, Cursor or any agent with connector support pushes what it already knows about your brand through Memo’s MCP connector. The screen shows the connector URL and a ready-made prompt with your kit’s ids baked in, and counts items as they land. See Connect your agent.
  • Set it up by hand — logos, colors, fonts, guidance and photography, one screen at a time.
The review fills in what a scan misses. Found no logo? Add logo uploads one, and it is saved as the kit’s primary logo. No written rules on the site? Add a rule takes a do or a don’t in your words. Both count toward the save like anything the scan found. After a website scan or brand-book import is saved, onboarding shows what landed as count chips — Colors, Fonts, Logos, Photos, Rules — and each opens its own screen. Back returns to the scan result, so you can keep adjusting. Fine-tune by hand walks all the manual screens with your imports already in place. Skip and Continue always work; you can import later from the kit page.

​
Scan your website

Paste a homepage address at the top of the kit’s empty state (or in onboarding) and Memo builds a proposed kit from the site itself:
  • Colors from the stylesheets. Every hex is read from the site’s CSS character-for-character, never from a model’s reading of a picture. A color takes the name the site gives it in its code (--brand-purple becomes “Brand purple”); a color the site does not name stays unnamed rather than getting a made-up name. Name it in review if you like. Memo ranks colors by how much the site uses them and by the site’s own token names (a --brand-500 beats a one-off gradient stop), collapses tint ramps into one color per hue, and keeps neutrals separate so a palette of grays can’t crowd out the accent.
  • Typefaces the site loads — @font-face files, Google Fonts links and named font variables count as strong evidence; system stacks (Helvetica, Segoe UI, -apple-system) are ignored. Each font shows an Aa sample in review when Memo can load it (Google Fonts), and you pick its job (heading, body, display…) from the menu beside it. When the site names a font but never loads it, the scan is not sure it is the brand font: the row says so, starts unticked, and you tick it to keep it.
  • The logo — the header’s inline SVG (rasterized, with the site’s CSS variables resolved), logo images, and the app icon as the symbol mark. Partner, customer and app-store badges are filtered out.
  • Imagery from the whole site. The homepage’s social preview and hero pictures, pictures that load as you scroll, and banners set as page backgrounds, plus the pictures on each page linked from the main menu — up to six sections, read in parallel. Pictures that repeat across pages (cross-sell cards, badges, a template hero) are treated as site chrome and skipped.
  • Collections named the way the site names them. Every section page becomes a proposed collection under its menu label — “USDA Certified Angus”, “Local Angus”, “Free-Range Chicken” for a butcher; “Women”, “Men”, “Sale” for a fashion store — holding the imagery found on that page. Homepage imagery files by type (Editorial, Product…) as before. Rename or uncheck any of them in review; saving creates them in the kit’s Collections tab.
  • Voice, guidance and rules — written from the full copy of every page read, not a homepage excerpt. The guidance document always has the same sections: About the brand, Audience, Voice & tone, Taglines & key phrases, Vocabulary, Visual language, Color usage, Typography and Imagery. Taglines are quoted word for word, and a section the site gives no evidence for says so instead of guessing. The do/don’t rules cover the whole offer, not just the homepage.
One fast vision call names and classifies what the parsers found (roughly 2–3 credits; up to about 6 if the backup model has to step in); it can only choose among the values the site actually declares. Some sites turn away any visit that is not a real browser (luxury houses and many large brands do); when that happens, Memo opens the page in a real browser and reads it from there, which takes about a minute longer. If the site still refuses, or is a pure JavaScript shell, the scan says so — upload a brand book or continue by hand.

​
Auto-extract from a brand book

If you already have a brand book PDF (or a brand-guidelines PNG / JPG), drop it onto the brand book dropzone in the kit’s empty state. Memo reads the whole document — cover to cover — and builds a proposed kit:
  • Colors, verified against the book. Printed hex codes are read from the PDF’s own text and double-checked, so what’s saved matches the page character-for-character. A value the book prints but that does not match what was read shows Check in review, so you know which one to look at.
  • Fonts with roles (heading, body, display, monospace, accent), weights, and a specimen crop.
  • Logos as variants, not repeats. Each colorway appears once, named the way the book names it (“black on yellow”, “light gray on dark”). Co-branding lockups are tagged as lockups, and repeated marks from other pages arrive collapsed behind a “show repeats” toggle.
  • Imagery grouped into collections. Photos and illustrations come pre-filed into named groups — Portraits, Product, Editorial, Illustrations — with short names, treatments, and subject tags. Rename a group or change an image’s type to move it; saving creates these as the kit’s photo collections.
  • Rules, ready to enforce. Every do/don’t from the book’s usage pages (“Don’t stretch the logo”) becomes its own rule, applied to every generation for this brand.
  • Brand guidance — the voice, usage, and photography direction distilled with page citations, shown on the kit’s Overview section by section, each part editable in place.
Everything lands in the review modal first — nothing is saved until you confirm. Each item carries a page chip (like p.12) that opens the brand book at its source page. Click Save and the kit is built: identity, collections, rules, and guidance in one pass. Extraction runs on Memo’s accuracy-tier model and is billed in credits at save-quality prices (a typical book runs a few hundred credits, shown before you start). If you cancel while it reads, the extraction stops straight away: you pay only for the reading steps that had already finished, and the kit is not marked as extracted. The dropzone accepts PDF, PNG, and JPG up to 20 MB. The brand book itself is stored on the kit, so you can re-open (or re-extract) it later without re-uploading. Connected agents can run the same intake over MCP with their own model — see Connect your agent.

​
Ambient vs @mention

Not all brand context loads automatically. The system distinguishes between:
  • Ambient — a lightweight asset inventory and active kits’ voice summaries are available to Studio chat. This is background context, not a selection of every kit’s colors, fonts, and images.
  • Selected kit or @mention — choose a kit in the chat’s Brand kit picker, or type @<kit name>, to load its colors, fonts, guidance, rules, and asset catalog. A limited selection of assets is shown as visual context; the entire library is not sent as images. The picker remembers the kit for that conversation; choose None for work without a selected kit.
Skills are @mention-only. They never run ambiently — you invoke them explicitly.

​
Multiple Brand Kits

Use separate kits for distinct brand identities, such as different clients or product lines with their own guidelines. Use Studio projects for individual pieces, and collections for grouping references. For a film or ad whose characters, places and look you reuse, make a Cinema Kit and combine it with the Brand Kit in Studio. A project can exist without a Brand Kit. Collections and projects are separate today; naming a collection after a project does not attach it automatically.

​
Tips

  • Start with assets, not text. Upload your best reference images and let analysis extract the palette and style. Write the Voice & Rules doc after.
  • Keep the Voice & Rules doc specific. “Use warm, directional lighting” beats “make it look good.” Banned words beat positive platitudes.
  • The voice summary is auto-maintained. You edit the full document; Memo summarizes it for ambient context so short prompts stay cheap and long prompts stay rich.