LLM Debate Council
Don't trust one AI.Make them debate.
Many AIs, one answer you can trust — fact-checked, with transparent reasoning. Built-in models and credits are included — start with no keys of your own.
Free plan · built-in models · credits included · or bring your own key — 0 credits
Watch the council argue it out
One question, three opinions, one decision. No editing.
Go Postgres. Transactions, JOINs, mature tooling — you won't shoot yourself in the foot early on.
Not so fast. If the schema is still fluid, rigid tables will slow down your first weeks of iteration.
Still Postgres — but keep the volatile parts in JSONB until the data model settles.
Start with PostgreSQL. Claude's caveat holds: while the schema is fluid, keep flexible fields in JSONB.
How it works
The council is already staffed with built-in models from different vendors. Want more control? Add your own keys and mix vendors.
Eleven thinking formats — from open debate to red-team and SWOT.
Agents discuss, fact-check, and converge on a single answer with transparent reasoning.
Your council by tier
Every plan ships with ready-made personas that debate for you. The model under the hood is hidden — you pick a way of thinking, not a brand.
Builds the strongest positive case
Hunts for risks and weak spots
Reasons from data and structure
Weighs trade-offs and second-order effects
Challenges the consensus
Gives the best-justified decision
Backs claims with facts and sources
Focuses on what will actually work
Attacks the strongest version of the plan
11 ways to think
One format per task. Choose how the council argues — from a fast brainstorm to a strict vote.
- ProDebate
Sides argue toward consensus — with agreement metrics and a stop on opinion delta.
many roundscostBest fortough calls — weigh pros and consHow it works- 1Each agent argues its own position
- 2Rounds of debate with agreement metrics
- 3Stop on opinion delta → verdict synthesis
- ProCouncil
Each answers alone, anonymously ranks the others — a chair picks the best.
a few roundscostBest forpick the best of independent answersHow it works- 1Each answers independently, blind to others
- 2They anonymously rate each other's answers
- 3A chair synthesizes the best answer
- StarterBrainstorm
Maximum ideas, no criticism; we stop when nothing new comes up.
1 passcostBest forgather many ideas fast - ProDevelop an idea
Take a single idea and deepen it from every angle.
a few roundscostBest fordevelop one idea in depth - ProStress Test
A defender against attackers, plus a dedicated red-team stress stage.
many roundscostBest forfind weak spots and risks - FreeRole analysis
Each agent plays a role: SWOT, six hats, stakeholders.
1 passcostBest forviews from roles: SWOT, stakeholders - StarterSocratic
One agent asks ever-sharper questions until the truth becomes clear.
a few roundscostBest forsharpen a vague, raw thought - ProSynthesis
Agents co-write one document, marking each change as added or fixed.
a few roundscostBest forturn the discussion into one document - FreeExpert panel
Experts weigh in from their own angle; a verifier reconciles the findings.
1 passcostBest forindependent expert opinions - ProEnsemble
Independent answers, iterated down to a single one.
many roundscostBest formerge several answers into one - FreeIn turn
Agents answer strictly in turn, building on what was already said.
1 passcostBest forstep by step — each builds on the last
Built for trust
The council doesn't just hand you an answer — it shows where it came from: with links, metrics, and built-in dissent.
Claims are marked VERIFIED / DISPUTED / FALSE with links to their sources.
Models reach the live web and ground answers in fresh data, not just memory.
One agent deliberately argues the weak side — a guard against false agreement.
Your documents become memory and are pulled into debates and chats automatically.
See how the decision was reached: every turn, the verdict, confidence, and dissent.
Confidence and disagreement scored at every step — no hidden assumptions.
The council pulls claims out of an answer and checks each one — here's how it looks.
Mount Everest is Earth's highest peak, standing about 8,849 metres.
VerifiedTraining a large AI model uses more energy than a small country does in a year.
DisputedThe Great Wall of China is visible from the Moon with the naked eye.
FalseIllustrative example. In the product, links point to real pages found on the web.
Why a multi-vendor council wins
One model is one opinion with one set of blind spots. Different companies' models have different strengths — debating them together catches what any single AI would miss.
Grok, GPT, Claude, Gemini, DeepSeek, Qwen — together
Built-in models by plan — or your own keys to control spend.
Confidence & agreement metrics on every turn
Transparent transcript — see how the decision was made
Plans & credits
Built-in models on our keys in every plan — they spend credits. Your own keys (BYOK) are optional: 0 credits.
* photo/document uploads work with agents whose model supports vision.
The Free plan's monthly limit resets on the 1st of the calendar month, not 30 days after signup.
On PRO, the council runs on 2026 flagship models
On the PRO plan the debate runs on the year's best models — from different vendors, each strong at something else. Their clash catches what a single AI misses.
Deep reasoning and long context — holds the thread of complex analysis.
Research and fresh facts — strong at search and up-to-date data.
Sharp opponent — hits weak spots and false consensus.
Strong logic and math — rigorous derivations and reasoning checks.
Topped-up credits don't expire; the monthly limit refreshes each period.
One-time packs, never expire, stack on top of the monthly limit.
+Credits or your own keys — what's the difference?
Credits power the built-in agents on our keys: nothing to set up, spent by the model's multiplier. Your own keys (BYOK) means you connect your provider key — the debate runs at 0 credits and you pay the provider directly for tokens. You can mix both: some agents built-in, some on your own keys.
We built a process that makes models prove their conclusions — not just state them confidently.
We are not trying to replace the best AI models. We create the conditions where the best AI models work together.
Plug in whatever you want: our models or your own keys
Built-in models already run on your plan's credits — no keys needed. Your own provider keys are an option: connect them, pay the provider directly for tokens, and spend 0 credits. The easiest way to start is Groq or OpenRouter — keys are free and instant.
console.groq.com — free instant key, very fast models.
openrouter.ai/keys — one key for dozens of models, some free.
platform.openai.com/api-keys — GPT models, billing required.
console.anthropic.com — Claude models, billing required.
aistudio.google.com/apikey — Gemini models, free tier available.
console.x.ai — Grok models by xAI.
platform.deepseek.com — strong, cheap DeepSeek models, paid.
alibabacloud.com — Qwen models via Model Studio, paid.
api.together.xyz — open models (Llama, Qwen, DeepSeek), paid.
fireworks.ai — fast open-model inference (Llama, Qwen), paid.
deepinfra.com/dash/api_keys — a wide catalog of models (Llama, Qwen, DeepSeek and more) at great prices, often cheaper than the original provider.
console.mistral.ai — Mistral and Mixtral models, paid.
Questions, answered
+Do I need my own API keys?
No. Built-in models run on your plan's credits — you can start without any keys. Your own provider keys are optional: then you pay the provider directly and spend 0 credits. The easiest key to try is Groq or OpenRouter — both are free.
+Which AI models are supported?
OpenAI, Anthropic (Claude), Google Gemini, xAI (Grok), DeepSeek, Groq, OpenRouter, Alibaba (Qwen), Together, Fireworks, Mistral, and any OpenAI-compatible endpoint.
+Is there a free plan?
Yes — 5 debates a month, 2 agents, 3 modes, all 15 languages. Paid plans add more agents, all modes, web search and fact-check.
+How is this different from ChatGPT?
It's not one model answering — it's several models from different vendors debating, fact-checking each other, and converging on one reasoned decision.
+What languages are supported?
The interface is in 15 languages, and agents reply in your chosen language.
+Can I export the result?
Yes — the Pro plan exports any debate to PDF or DOCX.
Make your next decision with a council, not a guess.
Free to start. Built-in models included. The difference shows in one debate.