Exec Room · by ANDRSAgents for your business, working together.
AI agents usually run hidden behind the scenes, in separate tools and scripts. In Exec Room they work together in the open, and a person approves every action.
“Agents that work together write better drafts, write better code and help you make better-rounded decisions for your business. My goal for Exec Room is to make all of that possible in a safe and controlled business environment.”
Agents you can see, actions you approve.
Your agents, in one room
They work together in the open instead of in separate tools and scripts. Requests go to one lead agent, who brings in the others when they're needed.
Nothing happens without your approval
Agents propose; you approve or ask for a revision. Makers write drafts, and nothing can post, publish or send on its own.
The team grows with you
Add advisors you consult or makers who produce drafts, or have the lead agent design a new one. Each new agent joins only after you approve it.
A room for every team
Exec, marketing, development and so on. Make a room yourself and pick the agents that team needs, or let ANDRS set it up for your organisation.
Each room only reaches what you allow
A new room starts with no extra folders, sites or connections. You add what its job needs in the room's settings, and every agent is told what's allowed.
What you need to run it yourself
macOS or Linux, Node.js 20+, and Claude Code. By default the room runs on your own Claude plan's usage. You can also start it on your own Anthropic API key, which is billed per use and stops at a spend limit per run and per day. AGPL-3.0 licensed.
What you tell the agents, and what you can look back on.
A brief per room
What every agent in the room should know about your business: customers, pricing, tone. You write it in the room's settings, in up to 3,000 characters. The lead agent, every advisor and every maker get it as background. Agents may quote it in replies and drafts, so it is not the place for secrets.
Every chat is kept
Each chat in a room is kept. Go back to one and give it a name. The lead agent picks up where the chat left off for as long as Claude Code keeps its own log of that chat, about 30 days after the last message. An older chat is history only, and the lead agent starts fresh there.
Every decision, listed
The room's settings list every proposal you approved, sent back or dismissed in that room, across all its chats: who decided, which agent proposed it, in which chat, when, and what came of it. The name under who decided is the owner's name. And the decision links back to the chat.
ANDRS's own room, up close.
This is the room ANDRS runs on today. A request lands with Coen, the COO agent, who brings in Theo (CTO), Priya (Head of Product) or Maren (CMO) when he needs them.
Coen proposes; you approve or ask for a revision. Makers like Vera (LinkedIn, X and Substack) only start work after your approval, and they write drafts, never posts. Nothing in the room can post, publish or send on its own.
Need another specialist? Add an advisor or a maker, or ask Coen to design one. It joins the room only after you approve it.
ANDRS's room is one example, and the team in it is ANDRS's own. Your room starts with one lead agent, Coen; you build the advisors and makers yourself, or ANDRS sets them up with you. You can make a room for any team, like marketing or development, and choose which agents are in it; one agent can work in every room that needs them. The open-source version is in private beta: join the waitlist to hear when the code is public, or to have ANDRS set it up for you.
Short answers, before you ask.
Is a person always in the loop?
For anything that changes something, yes. Agents read, think and answer on their own, within what the room allows. To get something done they propose it on a card: a job for a maker, a GitHub issue, a Notion page, an email draft. Nothing happens until you approve it, and you can send it back with a note or dismiss it.
Where does my data live?
On the computer that runs the room. Chats, proposals and the list of decisions are files in the room's own folder. The agents are Claude Code, on your own Claude plan or your own API key, so what you and they write goes to Anthropic the way any Claude Code session does. What you approve goes where you sent it: a draft file, GitHub or Notion.
What happens to files I attach?
You can attach up to three files to a message: text, CSV, PDF or images, 5 MB each. The agents working on that message can read them, and the room deletes them when the turn ends. What an agent read from a file stays in Claude Code's own log of the chat and in anything the agent quoted.
What can't it do?
It can't post, schedule, send messages or deploy: those actions don't exist in the room. Agents can't run commands on your computer, and they write only draft files. It is one owner's room on one computer, with no logins or seats, and it runs on macOS or Linux, not Windows. On your Claude plan it shows what each message used and sets no limit of its own. On an API key it shows what each message cost and stops at a spend limit per run and per day.
Who is behind it?
ANDRS: an AI consultancy and an in-house product studio, founded by Caroline Vrauwdeunt. ANDRS runs its own business in an Exec Room, the one in the example above. Exec Room is open source under the AGPL-3.0 licence.
How do I get it?
Exec Room is in private beta. Join the waitlist below and ANDRS will be in touch. You can set it up yourself, or ANDRS sets it up with you.
A room for every team. Make your own, or let ANDRS set it up.
Exec Room's open-source version is in private beta. Join the waitlist and we'll let you know when the code is public, or get in touch if you'd like ANDRS to set it up for your organisation.
You're on the list. Thank you.
We'll let you know when the open-source version is public, and get in touch if you'd like ANDRS to set up your rooms.