Gadaa Ask

Gadaa Ask sends your question to five independent frontier AIs — and delivers one verdict: what they agree on, how strongly, and where they split. Because a confident answer isn't the same as a correct one.

Ask the councilSee a real deliberation ↓
Featured on There's An AI For That

One AI vs. the council

The same question, asked two ways. Each deliberation below is a real record from the corpus, excerpted for length.

What one AI said
a single model

Absolutely. Authentic living is the master key — know yourself and freedom follows. Every great tradition says the same thing: self-knowledge is liberation.

“The master key,” said one model. The council split on whether that holds.

What the council found
Dissent

Authentic living is also not only taking what we have read or have been told to be the 100% all. Being able to see one’s truth based on one’s experience and intelligence.

Nemotron-3 UltraGLM-5.2Kimi K2.6Mistral Large 3Qwen3.5 397B

The claim you're making — that authentic living means testing what you've been told against what you've actually experienced and understood — is philosophically robust and worth taking seriously on its own terms before asking what it implies for AI.

On the human side, there is no real disagreement: all three lines of reasoning affirm that authentic living in this sense is not only possible but constitutive of what it means to be a person navigating the world. Received knowledge — from books, tradition, authority, culture — is a starting point, not a verdict. The person who never interrogates it again…

The council split · Methodological disputeFull record →

How it works

01

You ask once.

One question, the way you'd ask a colleague.

02

The council answers independently.

Five named seats, and none sees the others. Nothing is averaged and nothing is smoothed over — independence is what makes agreement mean something.

03

You get the verdict, and the receipts.

One readable answer, structured by confidence tier — how much agreement was behind it, and where it broke down.

The models come from different labs, trained on different data, and they fail in different ways. That's the point: when they agree, it means something. When they don't, you should know.

And the record talks back

Every deliberation comes with the Docent — a guide that answers your follow-up questions strictly from the record. It can walk you through a split. It will never settle one the council didn't.

The four tiers

Every answer carries one of four marks. Read the mark before the answer — it tells you how much agreement was behind it, and whether the run could carry a stronger one.

Strong Consensus

Every model agreed, including ones built by different labs on different data. That's rare. It still isn't proof.

High Confidence

The council agreed — but either a step short of ruling out a shared blind spot, or fully on a run whose grounding or roster could not carry the top mark.

Advisory

A lead, not a verdict — either the council did not converge on one position, or the run itself could not support a stronger mark.

Dissent

Models argued opposite sides of the same question. That's the finding, not a failure.

Dissent is a feature

When five independent AIs disagree, that's not a failure of the tool. That's the most important thing the tool can tell you. Anything that hides the split is choosing confidence over honesty.

Gadaa is not for everything. Here's the honest split.

Bring these questions
  • Decisions where being wrong is expensive — medical research directions, legal gray areas, financial and strategic calls you’ll have to defend.
  • Claims you need to verify before you repeat them — statistics, historical assertions, "I read somewhere that…"
  • Contested territory — questions where you suspect experts disagree and you want the disagreement mapped, not papered over.
  • Research synthesis where a single model’s blind spot could quietly shape your conclusion.
Skip the council for these
  • Drafting, rewriting, brainstorming, and creative work — a single model is faster and just as good.
  • Writing and debugging code — use Claude, GPT, or your IDE’s agent directly.
  • Quick lookups and casual conversation — deliberation takes longer and costs more than a chat message; that’s a bad trade for "what’s the capital of Australia."
  • Questions you want a specific answer to. The council reports what it finds, including "we disagree." If you want an AI that simply agrees with you, five of them are worse than one.

For teams: the deliberation record is the receipt — attach it to the decision, and the reasoning survives the meeting.

CouncilAnthropic (Chair) · Google · OpenAI · xAI · DeepSeek · Moonshot

A council you can name is a council you can hold accountable. Every seat is chosen for a different training lineage — they fail differently, on purpose.

Where it comes from

Gadaa is an Oromo system for arriving at truth that holds.

No single source sees all of it. An event, a position, a decision — there is always more than one angle, and the truth is more whole when more of them are accounted for. So the answer is reached in council, not handed down by one voice.

It carries a principle called Safuu: you meet every question the same way, with the same moral seriousness, and you stay accountable for the answer you land on.

That is why this is a council and not a model. But it asks something of you, too. The point is not to find the voice that already agrees with you — it is to be willing to hear the one that doesn’t.

Free to start — 50 lifetime credits, no card required. Core is for asking; Plus adds the frontier council, question by question.

Full pricing →

Ask something you'd want a second opinion on.

One question. Five seats. An honest verdict.

6,000 characters left