wagent-wiki
guide · shared ai memory for teams

Brand standards and client context in one shared memory

A small agency runs on brand guidelines and client context that every assistant should already know. Here is how to give your team's AI one shared knowledge base that stays current — and stays under your control.

Why every assistant needs the same client facts

A small agency runs on details that must stay consistent. Brand voice, logo and color rules, approved and banned claims, the client's product names, who signs off, and what not to touch. When one designer's Claude, one writer's ChatGPT, and one developer's Cursor each hold a different copy of those facts, the work drifts. One draft uses last quarter's tagline. Another names a feature the client just renamed.

Shared AI memory for teams stops the drift. Instead of a private memory locked inside each person's assistant, the team keeps one knowledge base that every assistant reads before it writes. Claude's built-in memory is per user, so it cannot do this on its own — the notes one person's Claude keeps never reach a teammate's.

How to give your team's AI one shared knowledge base

You do not need a heavy system. Four habits carry most of the value:

  • Write the standing facts once. Brand guidelines, tone rules, approved and banned phrases, and a short profile for each client. Use plain, short sentences an assistant can quote back.
  • Keep one source of truth per client. One focused space per client project, so context for Client A never bleeds into work for Client B.
  • Let the assistants read and write it. Connect each teammate's assistant to the same store, so it pulls the current rules and can add what it learns during the work.
  • Keep it current. When a client changes a name or approves a new claim, edit the shared page — not twenty private chats.

Client work needs governance, not just a shared doc

An agency holds confidential client context, so a shared store is only safe if you can control who sees what and prove what changed. Check for four controls before you trust any tool with client facts:

  • Roles that fit a studio. Owner, admin, and member, so a freelancer or a junior does not get the keys to every account.
  • Private by default. A client's space stays closed until you open it, and a sensitive wiki can be restricted to named members only.
  • A full record. Every read and write logged, so you can see which assistant touched which client's context, and when.
  • An undo you trust. Version history on every page and one-click revert, so a bad automated edit is a thirty-second fix, not an incident.

Where agent-wiki fits

agent-wiki is one honest way to run this: a neutral, cross-assistant store that every teammate's AI reads and writes over MCP, and that you can export as plain markdown whenever you want. It was built around these governance needs — owner, admin, and member roles, wikis private by default and restrictable to specific members, and a logged history with one-click revert. See how access and logging work, or read why we built it for the longer story. For the full feature set and plans, start at the overview.

Give brand standards and client context one home

Point every assistant at the same current source of truth, and stop re-explaining each client in every chat. Connect your first assistant and aim it at a shared wiki.