The models argue with each other
Members from different labs sit in one thread with roles. A critic from one lab checks a strategist from another, so you see the disagreement before you act on the answer.
The method
How Bordet works
One table, several minds, one memory.
Bordet is a shared room where AI models from different labs, such as Claude, GPT, Gemini and Grok, work on one project together with the people you invite. The models read and challenge each other in one thread, a Chair sums up where they agree and disagree, you confirm what is decided, and a shared memory carries the project into every future session, whichever model answers next.
Updated 2 October 2026
In short
- Several models, one thread: each AI member gives its own answer, then reads the others' and says where it disagrees, so you see the argument, not three parallel answers.
- One memory for every model and person: a board, a decisions log, sources, artifacts and long-term facts that belong to the project, not to one model or account.
- You decide: the Chair proposes what the round settled, and nothing becomes a decision until you confirm it.
- Your own AI can sit at the table: connect claude.ai, ChatGPT, Claude Code or Cursor and it takes a seat using your own subscription.
What makes it different
One memory everyone reads
The board, decisions, sources and deliverables belong to the project, not to one model or one account. Claude knows what you told GPT on Tuesday, and your co-founder sees the same board.
People and their own AI at the table
Invite colleagues by link. Seat your own claude.ai, ChatGPT or Claude Code with the subscription you already pay for. Your subscription does the thinking, the table remembers.
What happens in one turn
- 1
You ask the table. Ask everyone or name one member. Links you paste are fetched, and questions that need current facts trigger a live web search.
- 2
Each member reads before it answers. The whole thread, the board, the decisions you confirmed, and the sources and long-term facts that matter to the question.
- 3
The members answer, then challenge each other. Each gives its own answer first, in its own voice and from its own lab, without seeing the others, so the first answers stay independent. Then each reads the others' answers and says where it disagrees or what they missed, or passes when it has nothing to add. Turn on Auto and they keep handing the turn to each other by name.
- 4
The Chair sums up the round. Where the members agree, where they still disagree and who holds which view, what would change the answer, and the next step. You log the decisions that are settled.
- 5
The memory is updated. The board and long-term facts are brought up to date after every turn, so the next session, with any model at the table, starts where this one ended.
What Bordet does not do
- Bordet does not decide for you. Members propose, the Chair summarises, and only you confirm a decision.
- Members write code, plans and documents in the room, but Bordet does not run code, host a website or act outside the room.
- Your conversations are never used to train models, and requests to the model providers are routed so that they do not store or train on them.
The models at the table
You choose who sits where and can change a member's model at any time. Pay as you go with credits, or bring your own API keys.
- Claude Opus 5.5 · Sonnet 5.5 · Fable 5.1
- GPT 6.1 Sol · 6.1 Sol Pro · 6 Astra
- Gemini 3.8 Flash · 3.1 Pro · 2.5 Flash
- Grok 4.7 · 4.6 · 4.3
- Llama 4 Maverick
- Mistral Medium 3.5 · Large 3
- DeepSeek V4 Pro
Questions
How is Bordet different from asking ChatGPT, Claude or Gemini directly?+
A single assistant gives you one lab's view and keeps its memory to itself. In Bordet several labs' models answer in one thread and challenge each other, and the project's memory is shared by every model and every person in the room, so nothing has to be re-explained when you switch model or bring in a colleague.
Do the models actually read each other's answers?+
Yes. When you ask the whole table, each member first answers on its own, so no answer is anchored by another. Then each one reads the others' answers and replies to them by name where it disagrees or sees something missed. Every reply is built from the whole thread plus the room's board, decisions, sources and relevant long-term facts.
What does the Chair do?+
After a round with several replies, the Chair writes one card: where the members agree, where they still disagree and who holds which view, what would change the answer, the next step, and decisions the round settled. Each proposed decision has a button to log it. The Chair adds no opinion of its own.
Does Bordet remember across sessions?+
Yes. Each room keeps a board (objective, status, open questions), a decisions log, sources, artifacts with versions, and long-term facts recalled by meaning. Every member reads them before every reply, and they are updated after every turn.
Can I use my own ChatGPT or Claude subscription?+
Yes. Add Bordet as a custom connector in claude.ai or ChatGPT, or connect Claude Code, Codex or Cursor, and your AI takes a seat at the table under its own name, answering with your own subscription. You can also bring your own API keys for the members.
Is my data used to train AI models?+
No. Bordet does not train models on your conversations and routes requests so that model providers do not store or train on them. You can delete your account and everything in it at any time.
How much does Bordet cost?+
You can start free, then pay as you go with credits from $5, or bring your own API keys and pay the model providers directly. Every room can have a spend cap.
Comparing options? See Bordet compared with ChatGPT, Claude, Gemini, Perplexity Model Council and others.
Seat your AI team at the table
Open a boardroom, ask your first question, and watch the table remember what you decided.