Crads for teams

One company brain your whole team contributes to, an assistant on its own server for each person, and a company agent that holds the brain together. What goes where, who can change what, and what the owner owns.

A single assistant knows one person's work. A team needs two things at once: everyone's assistant should know what the company knows, and each person still needs a corner of their own. Crads for teams is built around that split.

What a team is#

Everything below lives in your accounts. Crads AI holds none of it.

• A company brain. A private repository in your GitHub. It holds who you are, how you sound, your policies and prices, shared client and process pages, shared skills, and what each person's assistant has published.
• A company agent. An assistant on its own server, and the only thing that can write the company brain. Every hour it merges what people published, with sources, and keeps a daily digest and a page for the week so far.
• An assistant for each person. Each on its own server, with its own brain, reading the company brain first.

Every server sits in your hosting account, made with your own API token. You, the owner, can open every one of them from your computer.

What flows up, and what never does#

Every night each person's assistant publishes facts from their work to the company brain: clients, projects, decisions. Every fact carries a line naming the person and the page it came from, so an answer can always say where it got something and who to check with.

Two things never leave a person's server: their personal/ folder, and any page marked private. That holds three separate ways: those pages are never copied into what gets published, the publishing job is blocked from reading them, and every publish is a commit you can inspect.

Be straight with your team about one thing: the company owns the servers. personal/ is private from the company brain and from colleagues. It is not secret from whoever administers the servers. It is for work in progress, not for private life.

Who can write what#

A person's assistant can only publish into its own repository, and the company agent copies in only that person's own folder. So no assistant can rewrite a policy, a shared skill or a colleague's pages, even by mistake.

How the rules change#

Policies, prices, processes, client pages and shared skills change only one way: someone proposes it, and the owner approves it.

• A person asks their assistant to propose a change. In Slack, anyone can ask the company agent the same way.
• The company agent puts it in Waiting for you on its Team page, with the exact lines that change and any warnings: a shared skill reaches every assistant on the team, so a skill that downloads things, mentions keys or tries to override instructions is flagged.
• You approve or turn it down. Nothing changes for anyone until you approve, and every decision is kept, with your note.

When people disagree#

Two people often know different things about the same client. The company brain does not pick a winner: it shows both, side by side, each with its source, and the digest lists every place it happens. An assistant asked about that client says that colleagues disagree and who to check with.

The company agent in Slack#

Connect Slack and anyone on the team can @mention the company agent, or message it directly, and get an answer from the company brain with its sources. It reads the company brain and nothing else, it ignores ordinary channel chatter, and it never answers in a channel shared with another company. Asked to change something, it drafts a proposal for you instead.

This is a different thing from connecting Slack to one person's assistant, which lets that assistant read and post as them. Both can be on at once.

A shared inbox#

If the team shares a mailbox, the company agent can read it every hour and add one line per new message to the company brain: who wrote, what they want, which client, and who knows them best. It is built so that reading outside email is all that part can do: it sees that one mailbox and nothing else, it cannot send or change mail, and what it writes is checked line by line before it reaches the brain. Anything personal or confidential is counted for the owner, not summarised.

When someone leaves#

You choose who is leaving and type their name. Then, in order: their personal/ folder is copied to your computer for you to hand back to them, their access to the company brain is removed, their repository is archived (kept, read-only), the company agent stops merging from them, and their server is deleted. What they already published stays with the company.

What it costs#

The software is free and open source. You pay the providers directly: one server per person plus one for the company agent, billed by your hosting company; Claude for each person (company seats or their own plans, your choice), billed by Anthropic; and Slack and GitHub as you already use them. Nothing is paid to Crads AI unless you want help setting it up.

Set up a team walks through it.