Skip to main content
Your Brand Book is what every agent reads before it writes a word about you. A page we draft, an outreach email, a listing fix and an answer from your public product feed all come from it. It holds:

​
Where a fact comes from

Each fact says how much to trust it:
  • drafted: we read it on a page (source_url says which). Nobody has checked it yet.
  • confirmed: you wrote it, or you approved it. Only confirmed facts go out under your name, for example in entity sync or your public product feed.

​
Public and private facts

Every fact is either public or private. A private fact is never in your public product feed and never in anything a public door says about you. Agents still use it in your work. A public fact goes out in public only once it is confirmed: a value we drafted is our guess, and it never goes out under your name. A new fact starts private, unless we drafted it from a page on your own site (your domain or one of its subdomains). Your site already says it in public, so the Brand Book does too. A fact drafted from anyone else’s page, such as a review site, starts private. You can make any fact public or private yourself. Making a fact private takes it out of your public product feed at the same moment. A fact your feed serves is always shown as public, never the other way round.

​
Who can change it

  • You, signed in. Your edit applies at once. A fact you write is confirmed.
  • An assistant or an API key (your own assistant over MCP, or a key you made). It can only propose. Its edit becomes a change of kind brand_book_edit in your Inbox, beside the value it would replace. Nothing changes until you decide.
The agents also propose Brand Book edits: a fact they found in a community thread, or one they read in a file you attached. A company profile edit from an assistant is a Brand Book edit too. All of these are brand_book_edit changes, and they wait for you the same way. Only a person decides a Brand Book edit. You approve or decline it signed in, in the app or on any door that carries your own signed-in session. An API key, an assistant’s connection or a shared review link cannot approve one or decline one, whatever its scopes. It gets a 403 that says where to decide. When you approve a fact, it is written as confirmed, by you. If you decline, the Brand Book stays as it was. A Brand Book edit never ships on its own, at any autonomy level. No Brand Book edit can change autonomy. Only you can raise a level, in the app.

​
The Brand Book page

In the app, the Brand Book row in the left rail opens your Brand Book as one page, at /brand-book. It shows, from top to bottom:
  • Facts: each fact with its value, whether it is drafted or confirmed, and whether it is public. A drafted fact links to the page it was drafted from.
  • Voice: your voice rules.
  • What we’ve learned: what your notes and declines taught the agents in this workspace. See below.
  • Claims allowed: the facts agents may say in public. These are the facts that are both confirmed and public.
  • Never say: your never-say phrases.
  • Competitors and Markets. Markets link to Settings, where you change them.
  • Connections: a link to where you manage your connected accounts.
  • Autonomy: each lane’s level, read only. It links to the Autonomy settings, the one place a person changes a level.
Edits waiting for your yes show inline. Each one sits under the fact it would change. It shows that fact as it is now beside what it would become, links the page or thread it was read from, and offers Accept and Decline. One tap decides it. The page shows three kinds of proposal:
  • An agent’s or an assistant’s Brand Book edit, such as a fact from a community thread or from a file you attached.
  • A page that says otherwise: a third-party page that decides your answers and contradicts one of your confirmed facts. Accepting it asks that page to match your Brand Book. Your Brand Book itself does not change.
  • A never-say or voice edit, which shows under Never say or Voice. A proposed fact your Brand Book does not have yet shows under New facts.
Accept and Decline do exactly what approving and declining the change in your Inbox does. A Brand Book edit that changed since you opened the page is not applied. The page tells you and shows the current version. To edit a fact yourself, press Edit under it and save. Your edit applies at once and the fact is confirmed by you. The page then says that entity sync has started and shows where each profile stands. Make public and Make private set a fact’s public flag the same way. If the Brand Book can’t be read, the page keeps showing the last version it read, with the time it was read, and tries again on its own. The first retry is after 5 seconds, and the wait doubles to at most a minute. If it has never been read in this browser, the page says so instead of showing nothing.

​
What we’ve learned

Under Voice, What we’ve learned lists what your workspace taught the agents, and what shapes their drafts now. See how the agents learn from you.
  • Learned rules: each regenerate note you kept as a standing rule. It shows the rule, its never-say phrases, and where it came from: a link to the change you wrote the note on, and the day you kept it. A rule you rewrote here says Edited by you. A kept note is listed here and not again under Voice.
  • Paused playbooks: each playbook your declines paused, with its reason, until when, and links to the declined changes that paused it.
Rules you removed and pauses that ended or were lifted are not listed here. GET /v1/learned-rules keeps the full history. Each row takes one tap:
  • Edit a learned rule, change its words and press Save. Every draft after that is checked against your rewrite, and its never-say phrases are read again from the new words. A rewrite can only restrict: one that would loosen or override the rules is refused, the same as a note. To loosen a rule, remove it.
  • Remove a learned rule. Drafts stop being checked against it, starting with the next one.
  • Lift the pause on a playbook. The agents may use it again, starting with the next run.
These are Brand Book edits like any other. From you, signed in, they apply at once. From an assistant or an API key, each is a brand_book_edit change in your Inbox, and nothing changes until you approve it.

​
Never-say phrases

A phrase you add to never-say stops the next draft that contains it, including a section you ask the agent to write again. Matching ignores case and works on whole words. We also suggest a few common filler phrases. A suggestion stops nothing until you add it to your own list.

​
Reading and editing it over the API and MCP

The Brand Book is one object on every door. Both reads return the same JSON. An edit carries exactly one of these:
A market_fact edit names a market the workspace reads, with language when that market has one. See per-market Brand Book values. From a signed-in person, the answer is status: "applied" with the Brand Book as it now stands. From a key or a connection, it is status: "proposed" with the change’s id and Inbox ref, and nothing has changed. A person then approves or declines it in the Inbox. A key needs the configure scope to propose an edit. configure with kind: "brand_book" is not yet offered to assistants over MCP. Until it is, an assistant proposes the same edit with POST /v1/config/brand_book, on https://api.attensira.com or https://webapp.attensira.com with the same key, or with PATCH /v1/brand-book, which only https://webapp.attensira.com serves for now. See the endpoints.