Skip to main content
Most teams explain their brand over and over: to a freelancer, to a new hire, to whichever AI tool they opened this morning. The Brain is where that explanation lives instead. You teach it once, and it reaches your agents, your teammates, and assistants outside Orisu. Find it at Brain in the sidebar.

Filling it

Click the + at the centre of the map to start encoding. “Encoding” is what Orisu calls taking something in and turning it into knowledge. From a website. Point it at your site. Orisu crawls it, reads your palette, fonts, voice and claims, and harvests your product imagery, so your real photography is available as reference material rather than something you re-upload each time. While it reads, a tray at the bottom of the map shows which step it is on and how many facts and images it has found so far. It is still there when you come back to the page, and when the read ends it says whether it finished, finished with a gap, or failed, with a way to try again. Upload anything. Drop in a brand book, a tone-of-voice doc, a past campaign, a spreadsheet, or images. Documents are read page by page and become facts that cite the page they came from. Images become brand media straight away. Bring over a brand you have already described. Most teams have explained their brand to an assistant before, at length. Import memory gives you a prompt to paste into ChatGPT or Claude, asking it to gather what it knows about your brand from your saved memories, past conversations and anything you uploaded. Paste its reply back and Orisu reads the brand out of it. You see what is in the reply before anything is saved: how many memories, which categories, how many word rules. A profile field you have already written is never overwritten, and the words you have banned are kept and merged rather than replaced. Just talk to the copilot. When you tell it something durable about your brand, it saves that to the Brain and shows you a chip on the message. You do not have to remember to file anything.

Reviewing what it learned

Anything Orisu worked out on its own arrives unconfirmed and waits for a person. The queue is at Needs review, with the facts Orisu is most confident about first, so the clearest calls are the quickest to make. When a read lands many facts at once, Confirm all and Dismiss all at the top of the list act on every fact waiting there. Dismissing never deletes anything: the facts move to history. Confirm what is right, dismiss what is not, and correct what is nearly right. Most of what Orisu works out on its own is close but not exact, so Edit lets you fix the wording or move it to a better category. Saving a correction confirms it at the same time, because a person wrote the words. Dismissed facts are kept in history rather than deleted, so you can see what the Brain used to believe and when that changed. Every memory shows who taught it and who signed off, so a rule you disagree with has someone you can ask about it.

The brand profile

Click your brand’s name at the top of the map to open its profile. It is split into five tabs:
  • Identity: what you make, and what you stand for.
  • Look: your palette and fonts. Give each colour a role (Brand, Accent, Dark or Light) and each font a job (Headline or Body) and a weight, so a generation knows which colour leads and which typeface sets the headline.
  • Voice: how the brand sounds, a few tone words, how it sounds on each channel, and how to say names people get wrong. Switch on Keep for a name and your agents will never translate it.
  • Rules: the words below, and your guidelines.
  • Market: where you sell, who you sell to, and your products with the selling points a generation is allowed to claim.
Each change saves as you make it, and your agents use it from their next run. On the map, every part of the profile you have filled in shows as a small square beside its category. Click one to jump to that tab. When you upload a brand book, the palette, fonts and voice Orisu reads from it arrive in the review queue like any other fact. Confirming one also adds it to your profile.

Questions

When the Brain is missing something, a questions chip appears at the top of the map, showing how many it has. Open it to answer them one at a time, or skip any you would rather leave. An answer goes straight into your profile, or becomes a confirmed memory. The Brain only asks about gaps. Once your website read, a document or a teammate has filled something in, the question about it goes away. The copilot can ask too, at a natural pause in a conversation, and never more than two per conversation. Answer in your own words and it files the answer for you.

Words your brand never uses

The profile has a Words section for terms you insist on or refuse. Mark one Never and your agents will not write it. Mark one Preferred and they will reach for it instead. Careful is for a word that needs a specific claim behind it. This is worth using even though your guidelines already say it in prose. A banned word is the most checkable rule a brand has, and a model follows “never write cheap” far more reliably than a paragraph that mentions it in passing.

Memories that go out of date

Some knowledge has a shelf life. Where you advertise this quarter, what the team approved last month and which campaign is running will all change, and an agent follows an old one as confidently as a current one. So memories in those categories carry a date. A card says “Expires in ~90 days”, and once the date passes, “Expired”. An expired memory is not deleted: it stays in your lists and on the map, faded, but your agents stop using it until someone clicks Still true, which starts the clock again. Reading the same claim off your website again renews it too. Channel and feedback memories last about 90 days, campaigns about six months, and products and audiences about a year. What your brand is, and its standing rules, never expire.

Reading it

Three views, same knowledge:
  • Map: your brand as a graph, with categories radiating from the centre and your media as thumbnails. Good for seeing the shape of what you know, and the gaps.
  • Memories: everything in a searchable, filterable list. Good for finding one rule.
  • Media: every image the Brain holds, at a size you can judge. Remove all clears the whole list after asking once. Search first and it removes only the matches, so you can drop all the logos, say, and keep the product shots. Images that came from your website come back if you read it again.

Using it

In the copilot. On by default. The copilot starts each message knowing your brand’s identity and your word rules, and can search the rest of your memories when a question needs them. So asking it for launch copy gets you copy in your voice, without you pasting your guidelines in first. The Brain switch beside the composer turns it off for a conversation where you would rather work without it. On the Create desk. Create has a Brain chip on the instrument row. Switch it on and the prompt you type is rewritten in your brand before the shot: your palette named in words, your materials and settings, your products where they fit, and none of the words you never use. The rewrite draws on the memories nearest to the prompt. Shot notes tell you it will happen, and the finished row keeps what you typed with the rewrite beneath it, so a reroll starts from your words. A rewrite costs a few credits on top of the generation, shown on the button. Outside Orisu. With the MCP server connected, your own Claude or ChatGPT can read the Brain, search it, and add to it. Anything an outside assistant proposes lands in your review queue like everything else. Wherever the Brain is searched (the copilot, an outside assistant, a Brain node in a workflow, the Create desk), it matches exact names as well as meaning. Ask about “Loomora Air” and you get the memory that names it, not just the ones that sound similar.

One brand per workspace

A workspace holds one brand. That is deliberate: if a workspace held two, an agent generating for one could be handed the other’s rules. If you work across several brands, say an agency or a house of labels, give each one its own workspace. Reading the same website again refreshes what the Brain learned from it and keeps every fact, image and agent already attached.