--- title: Knowledge description: BRAND.md, company facts and voice documents: the written half of your brand, as the Markdown files agents read. --- > Source: https://www.heyarchetype.com/docs/brand/knowledge Knowledge is the written half of your brand: what the company is, the facts agents may state, and how the brand speaks. It's a set of Markdown files, and BRAND.md is the one every agent reads first. Open **Knowledge** in the sidebar. The file list on the left has three groups: | Group | File names | What goes in it | | --- | --- | --- | | **Start here** | `BRAND.md` | The overview, the rules every task follows, and the routing that tells an agent what each kind of work needs. | | **Company** | `company/….md` | Facts about the business: who you are, what you sell, your prices, your positioning. | | **Voice** | `voice/….md` | How the brand speaks: messaging, voice and tone. | ## The starter files Every brand starts with BRAND.md, **About** and **Positioning** under Company, and **Messaging** and **Voice & Tone** under Voice. Each one opens as a skeleton with headings and prompts to fill in. Until you edit a starter file, it's marked **Starter text** and agents read it as a skeleton, not as the brand's word. Write over it to make it count. Each fact should live in one file. About holds the facts and the only prices, Positioning the strategy, Messaging the words, and Voice & Tone how they sound. Other files point at them instead of repeating them, so an update never leaves a stale copy behind. ## BRAND.md BRAND.md has three parts: - One paragraph on what the company is, who it serves and what makes it different. - **Always**: rules every task follows. - **Routing**: tables that match a task or a format (for example "Instagram story") to what it needs, split into **Required** and **Consider**. In the routing tables, name what a row needs as a reference, such as `[[template:Launch story]]` or `[[doc:About]]`. When an agent asks for help with a task, Archetype matches it against these rows and hands over exactly those items. Two buttons in BRAND.md's header show what agents get: - The token count (for example **≈1,200 tokens**) opens **What agents read first**: BRAND.md as it opens the brief, with every reference turned into the call that fetches it. - **Try a task** lets you type a task, such as "An Instagram story selling our newest product", and press **Get brief** to see the brief an agent would get for it. ## References Inside any Knowledge file you can point at something the brand holds with `[[kind:Name]]`. The kinds are `doc`, `template`, `component`, `artifact`, `logo`, `photo`, `color` and `type`. On the page, each reference shows as a chip. A reference to something the brand doesn't hold is flagged as not in the brand yet, so you can fix the name or add the item. Agents get each reference as the call that fetches the real item. A routing row Claude proposed and nobody has reviewed yet carries `[[drafted]]` and shows as **Drafted by Claude · review**. ## Write and organize files - To edit a file, press the pencil (**Edit Markdown**), write, then press **Save**. **Cancel** discards your changes. - To add a file, press **+** beside a group and choose **Blank file** or **Upload Markdown files** (.md or .txt, up to 2 MB each). - Drag files to reorder them within a group. - The **…** menu on a file offers **Rename**, **History**, **Download**, **Copy** and **Delete**. Each file also has a status and a comments thread. Only files that are **Complete** or have no status are served to agents. See [Item states](/docs/workspace/states). ## Ask Claude to revise Press **Edit** in a file's header to open a conversation on the right, and ask for a change, for example "make the tone warmer and cut the jargon". Claude reads every Knowledge file and can change several at once, or propose a new one. The conversation runs on desktop. When the proposal is ready, review it under **Proposed changes**: - Untick any file you don't want changed. - Edit Claude's wording directly. Your edits are marked **Your wording**, and you can revert to Claude's version. - Press **Apply to** (for example "Apply to 2 documents") to write the changes, or **Discard** to throw the proposal away. Nothing changes until you apply. Each request uses AI credits. See [AI credits](/docs/billing/credits). ## How agents use Knowledge The brief an agent gets when it starts opens with BRAND.md: your overview, your rules and your routing, each item with the call that fetches it. Agents can then list every Knowledge file, fetch one by its slug, or fetch them all at once. They're told to treat these files as the source for any copy they write. You can also download BRAND.md and the other files to use alongside your code. See [DESIGN.md, BRAND.md and skills](/docs/agents/files). ## Next steps :::cards - [DESIGN.md, BRAND.md and skills](/docs/agents/files): Use your Knowledge files outside Archetype. - [Templates](/docs/brand/templates): The layouts your routing points at. - [MCP tools](/docs/agents/tools): How agents read the brief. :::