← All sections

01 · Brand Layer

Start here

Three steps to your first on-brand asset, then the six decisions that give the same thing to a whole team.

01 · Your first asset

Your first asset

Three steps, once. After that a sentence is enough.
  1. 01

    Give your tool the brand

    Once per tool

    If your tool takes a remote MCP connector (Claude, ChatGPT developer mode, the coding agents), point it at the endpoint. If it does not (the Gemini app, ChatGPT Free, most locked-down workspaces), download one file and add it as Knowledge to a Gem or a Custom GPT. If it can neither connect nor fetch, or you need the logo and font files themselves, download the offline package and give it the folder. Every path gives the agent the same brand.

    MCP endpointhttps://galton-ai-assets.vercel.app/api/mcp
    Setup guide for your tool
  2. 02

    Paste the connect-and-remember prompt

    Once per project or Gem

    It tells the tool where the brand lives, that the brand is binding for the rest of the project, and that it should work in guided mode. In a Gem or Custom GPT it goes in the instructions field, so it keeps applying without you repeating it.

    Copy the prompt
  3. 03

    Ask in one sentence

    Every time after that

    That is a complete brief. Guided mode fills the surface, format, logo variant, icons and language from the GALTON defaults, asks at most one thing it cannot infer, then tells you what it chose and how to change any of it in one word.

    Try thisMake me a poster for GALTON, black version.
    More prompts, and how guided mode works

02 · Roll it out to a team

Roll it out to a team

Six decisions, answered for each size. What changes with headcount is not the technology, it is who decides and who supports.

The one rule · pick a single path and set it up centrally. Everything below follows from that.

Decision5 to 10 peopleOne team, one owner. The risk is not chaos, it is that nobody starts.30 to 50 peopleSeveral teams, nobody can train everyone. The risk is fragmentation.
Who sets it upThe owner, once, for everyone. On Claude Team an Owner adds the connector; otherwise the owner builds one shared Gem, or a Custom GPT if the workspace is Business, Enterprise or Edu.IT, once, for everyone. Never per person: 50 people configuring their own tools is 50 configurations and no way to tell which output came from which.
What people connect toOne path for the whole team, chosen by the owner.One path, approved centrally, plus a named fallback for locked-down laptops: the uploaded file, or the offline package where even that is blocked.
How people learn itOne 60-minute session: 10 minutes on what it is, 20 minutes watching three real outputs, 30 minutes where everyone makes their own.A 20-minute recording plus this page. The one live session is for the champions only; they run the 60-minute session for their own team.
Who answers questionsThe owner.One champion per team or market, first line. The owner only for exceptions and for changes to the brand itself.
Who approves whatInternal work: nobody. Anything leaving the company: the owner.Internal: nobody. Client-facing or public: the champion. Campaign or identity level: the owner.
What to watchHow many people made something in the first two weeks, and how often the owner has to correct an output.Those two, plus how many shared setups are still on an old brand version after a release. The brand.json surface always carries the current one.

Send people the tested prompts after the session, and the before and after page when someone asks what difference it makes.

03 · Keep it current

Keep it current

The brand will change. What that means depends on how you connected to it.
  • A connected tool is never stale. The MCP server reads the brand on every call. Saved copies of the orchestrator skill carry a version: compare it with https://galton-ai-assets.vercel.app/brand.json and re-fetch when they differ.
  • An uploaded file or an unzipped package is. A Gem or Custom GPT holds the copy of llms-full.txt it was given, and an unzipped package is the brand as it stood on the day it was downloaded. Both carry the version they were built from, so the check is quick, but someone has to replace them when it changes. That job belongs to a named person.
  • The changelog says whether you need to. Every Brand Layer release lists what changed. A minor bump usually means new knowledge for agents; a patch rarely needs anyone to do anything.
Read the changelog
Next →For AI agents