Roast Room is a free, open-source Claude Code skill. You type a business decision, and seven AI advisors argue about it, score the options and vote. You get a verdict and a PDF report. It was built by Nate Herk's AI Automation Society and is published on GitHub under the MIT license. Here is how to set it up.
What you need
Claude Code, Anthropic's coding agent. It runs on a paid Claude plan such as Pro or Max (Anthropic support). The repo says no API key is needed.
Node.js 20 or newer and Git on your computer.
The repo: github.com/nateherkai/roast-room, free under the MIT license.
Step by step
Who sits on the council. Each seat is a Claude sub-agent with its own "brain", a folder of notes built from that person's public books, letters, talks and videos. These are AI personas, not the real people. The repo says they don't speak for them, and none of them is affiliated with or endorses the project.
Jeff Bezos persona: Strategy, and how reversible the bet is
Paul Graham persona: Product, demand and startups
Alex Hormozi persona: Offers, pricing and sales
Gary Vaynerchuk persona: Marketing and attention
Dan Martell persona: People, hiring and founder time
Gino Wickman persona: Operations and scaling
Warren Buffett persona: Money, capital and risk
Each brain has files for frameworks, principles, positions, stories, voice and sources. A MindStudio write-up says the Hormozi brain alone draws on around 124 sources and roughly 1.3 million words of material (MindStudio).
How to install and run it.
Clone the repo. Open a terminal and run:
git clone https://github.com/nateherkai/roast-room.git
cd roast-roomStart the visual room (optional). This opens a small local web app that shows the advisors around a table:
npm startThen open http://localhost:4730 in your browser. The repo says the room only plays back session files and makes no AI calls itself.
Open Claude Code in the same folder:
claudeRun the council. Type
/roast-roomfollowed by your decision. For example:
/roast-room Should I launch a ₹999 cohort course for my AI tools audience, or keep growing my free content for six more months before selling anything? I have 40,000 followers, about 2% reply to stories, and I can spend 10 hours a week on this.Answer the Chair's questions. The council works better with real numbers: budget, audience size, time, what you've already tried.
Read the verdict. You get a short verdict in Claude Code, a full report saved under
sessions/, and a Download PDF button in the visual room.
Lighter modes from the README:
/roast-room --quick Should I raise my price from ₹499 to ₹799?
/roast-room --seats hormozi,buffett Should I take a loan to buy inventory for Diwali?
/roast-room ask hormozi How should I package my offer?--quick seats four advisors for two rounds. --seats picks only the advisors you name. ask gets one answer from one seat.
No terminal? You can also open Claude Code anywhere and ask it to clone the repo and start the room for you.
Good to know
It uses a lot of your plan. The README says a full council is 21 sub-agent runs. Use
--quickwhile you're learning.API mode costs money. If you set
COUNCIL_RUNNER=apiwith an Anthropic API key, it bills your API account, at about 25 calls per full council.They're simulations. The advisors are AI personas built from public material. They can be confidently wrong, and they don't know your market the way you do. Use the verdict as a checklist of questions, not as a decision.
Don't quote the personas as the real people. If you share outputs, make it clear they came from AI personas and aren't quotes from Bezos, Buffett or anyone else.
Check the code before you run it. It's open source, so read the README and skim the files before you give any tool access to your machine.
Sources
Found this useful? Follow @kevinsingh.me on Instagram for a new AI tool every day, and subscribe so you don't miss the next one.