Hard questions deserve more than one opinion.
Witan is an LLM council. It asks a panel of frontier large language models, GPT, Claude, Gemini, Grok, and Deepseek, the same question at the same time. You see where they agree, where they argue, and what each one believes alone. The points two models contradict each other on are usually the points worth thinking about yourself.
Witan is free. You bring your own OpenRouter API key. It stays in your browser. No signup required to try.
How it works
Three steps from prompt to verdict.
Pick your council.
Choose which models deliberate. The default is four strong frontier models from OpenAI, Anthropic, Google, and xAI. Swap in others whenever you want. Free open-weight models are listed too if you want to keep costs at zero.
Ask your question.
Attach a PDF, paste a draft, or just type. Every counselor sees the same context. Answers stream in parallel, one tab per model, so nothing is hidden behind a single voice.
Read the verdict.
A synthesizer pulls every counselor's response into one view. You get a headline, the points they all agree on, the topics they argue over, and the unique takes a single model held alone.
What you’ll see
A council verdict, in one view.
Your question
Should we A/B test the new pricing page before launch?
Headline
Yes, but split traffic carefully. The downside of a test is bounded; the downside of a bad pricing change is not.
Concord · 3 of 4 counselors agree
- Pricing changes carry real revenue risk. A test bounds the blast radius if the new page underperforms.
- Two weeks is the typical minimum runtime for a meaningful pricing signal at most B2B traffic volumes.
- Segment the readout by acquisition channel. Reactions vary more by source than they do across overall traffic.
Discord · 1 topic
Traffic allocation between control and variant
50/50 split. Maximizes statistical power, shortens runtime, and gives a clean readout.
GPT-5.5 · Claude Opus 4.7
10/90 toward control. Limits revenue exposure if the variant turns out to be worse than the current page.
Gemini 3.1 Pro
Singular · 1
Grok 4.20
Test the price display format independently from the actual price levels. Confounding the two is a common mistake that ruins the readout.
Every council run produces this view in about thirty seconds, plus each counselor’s full response one tab away.
Why a council
Why one model is not enough.
Frontier models sound just as confident when they are right as when they are halfway through inventing a citation. Ask only one and you cannot tell the two states apart.
A council fixes this from the outside. When several models independently land on the same answer, that is real evidence the answer is solid. When they split, that is real evidence the question is worth a second look. Witan surfaces both signals, instead of hiding the disagreement to look smart.
Pricing
Free forever. You pay OpenRouter, not us.
Witan itself is free. There is no subscription, no trial that expires, no usage cap, and no premium tier. You bring your own OpenRouter API key. The models you call are billed to your OpenRouter account at OpenRouter’s listed prices. We do not mark up, proxy, or rebill anything.
Your key lives only in your browser. Witan servers never see it, and your prompts go straight from your machine to OpenRouter. If you delete Witan from your bookmarks tomorrow, nothing about your billing changes.
Witan
$0
forever
No account required to try. Sign in only when you want history synced across devices.
A typical council run
≈ 3 to 10¢
per question
Four frontier models plus a synthesizer, billed to your OpenRouter account at cost.
Free open-weight models
$0
per question
GPT-OSS, Nemotron, GLM, and others. Use them when you want to keep the bill at zero.
What you get
Everything in the box.
Frontier models from every major lab.
OpenAI, Anthropic, Google, xAI, and Deepseek, callable in any combination.
Free open-weight models too.
GPT-OSS, Nemotron, GLM. Use them when you want to keep costs at zero.
Streamed answers in parallel.
One tab per counselor. Watch every response come in as it is generated.
Attachments that all models see.
PDFs, images, spreadsheets. Every counselor gets the same context.
Synthesized verdict.
Consensus, dissent on specific points, and the singular take each model held alone.
Synced history when signed in.
Conversations follow you across devices. Anonymous use stays in your browser only.
Branded PDF and Markdown export.
Take a verdict with you. Share it with a colleague. Bring it to a meeting.
Your key, your browser.
Your OpenRouter API key is stored in localStorage. It is sent directly from your browser to OpenRouter, never through us.
Ready to convene?
Try one council on a question you are sitting on right now. No signup needed.
Open the council