How it works
From your first answer to a decision
Eight stages. You stay in control at the two that matter: validating the requirements and answering the questions that change the outcome.
Interview
An AI interviewer asks about goals, budget, team, existing tools, constraints and priorities.
Requirements
You review the requirements sheet and validate it before any design starts.
Blind proposals
Each debater designs a solution without seeing the others, so nobody anchors on the first idea.
Critiques
Debaters read each other's plans and write critiques rated blocking, major or minor.
Your answer
When your answer would change the outcome, the debate pauses and asks you.
Revisions
Each plan is revised against the critiques it received, or defended with reasons.
Verdict
An independent arbiter rules on each option, scores them against your criteria and raises any angle nobody examined.
Report
You get a decision report to read, share or hand to your coding agents.
Debate replay
Watch the arguments land
A short replay of a textbook case (a chain of 12 clothing shops that wants to sell online), then its decision matrix. Move the weights and the ranking follows, as it does in the app.
Decision matrix
| Option | Feasibility | Score |
|---|---|---|
| Rank 1Turnkey platform | Feasible | 4.3 |
| Rank 2Open-source shop | Feasible with conditions | 2.8 |
| Rank 3Custom development | Not feasible | Excluded |
Sample data for illustration. In the app, the criteria and weights come from your interview.
Four modes
Pick the kind of debate you need
A full comparison for a big decision, a quick round for a small one, a stress test of the plan you already have, or an audit of the quote you received.
Full debate
Compare solutions
Up to four debaters each design an option without seeing the others, then criticise each other's plans and revise their own. The arbiter rules on each one.
A few questions
Express
One set of interview questions, then a single short round. A clear answer in a few minutes.
Your solution
Challenge my idea
Describe the solution you have in mind. The models hunt its flaws and tell you what would make it hold.
A provider's quote
Audit a document
Attach the quote, statement of work or architecture a provider sent you, and say what it is for. The models check it against your need: what is missing, what is vague, whether the days and the price hold up, and the questions to send back.
Example: a web agency quotes €38,000 for a booking website. Is it consistent? What is missing?
Ready-made templates
- Web app
- Mobile app
- Data and AI
- Migration
- E-commerce
The debate can read
- PDF spec
- Word document
- Web page
- GitHub repository
The deliverable
Anatomy of a decision report
The arbiter does not just name a winner. It shows its reasoning, so you can check it, share it and act on it.
- 1
Recommendation
The option to build and the conditions that come with it.
- 2
Interactive decision matrix
Each option scored against your priorities. Re-weight the criteria and watch the ranking change.
- 3
Debate map
Where the debaters agree, where they are divided and where they are opposed.
- 4
Risks and mitigations
What could go wrong, how likely it is, and what to do about it.
- 5
Cost per option
Build cost and monthly running cost for each option, marked as estimates.
- 6
Architecture diagram
A generated diagram of the recommended option.
- 7
Roadmap and fact check
Phases and milestones, and every statement marked verified, estimated or assumed.
Debate map
Each topic placed by how far the debaters are apart.
Agreement
- Payments handled by the platform
- Launch before the holidays
Divided
- Commissions on sales
- Real-time stock sync
- Hosting and maintenance by the agency
Opposed
- Custom development
Rewritten for your reader
CTO
Trade-offs, architecture and risks in full.
Executive
Decision, cost and timeline on one page.
Non-technical client
Plain words, no jargon, what it means for them.
Export it your way
- Markdown
- Word
- Architecture decision record
- Hand-off pack for coding agents
For coding agents
Debate before your agent writes code
Quorumind runs as an MCP server, so Claude Code, Codex or Cursor can start a debate before they build a feature. The verdict comes back as a plan they can follow.
- A hand-off pack with CLAUDE.md, AGENTS.md and PLAN.md
- Decisions recorded as ADRs in your repository
- Works with any MCP client
Before adding sign-in, run a Quorumind debate: build it or use a provider?
Debate started: 3 debaters, 1 devil's advocate.
Verdict: use a managed provider, feasible with conditions.
Wrote PLAN.md, AGENTS.md and docs/adr/0007-authentication.md
Features
Everything a hard decision needs
Built for decisions you will have to explain to a client, a manager or your future self.
Live debate
Follow proposals, critiques and revisions as they are written.
Expert personas
Cloud architect, security, cost control, UX or data-protection lawyer, up to four at once.
Devil's advocate
An optional voice whose only job is to attack every option, the favourite first.
Consensus check
The arbiter flags a too-easy agreement and raises the angle nobody examined.
Your context attached
Add a PDF or Word spec, a web page or a GitHub repository.
Framing templates
Start from a web app, a mobile app, data and AI, a migration or an online shop.
Capped costs
In the cloud, you pay the debate's fixed price, never more. Self-hosted, you see the cost of each role as it runs, up to a limit you set.
A different AI for each role
Claude, OpenAI, Gemini or Mistral, chosen for each debater and the arbiter, and Ollama models when self-hosted.
Ten languages
Debate and report in your language. Arabic is displayed right to left.
E-mail notifications
Ask for an e-mail when the debate needs your answer or the report is ready.
Branded reports
Put your name, your logo and your colour on the reports you export.
Open source
AGPL v3. Read the code, run it, change it.