Models debate, critique, and synthesize — producing answers more balanced, more nuanced, and more complete than any single model alone. GPT-4o, Claude, Gemini, and more deliberate on your behalf.
Every question goes through a multi-stage deliberation pipeline. Think of it like a panel of experts debating before giving you a final answer.
3 or more AI models (from OpenAI, Anthropic, Google, xAI) each answer your question independently, without seeing each other's work. This ensures diverse perspectives.
Each model reads and scores the others' responses, identifying factual errors, logical gaps, missing perspectives, and strengths. This peer-review stage catches mistakes no single model would find.
A senior model reads all responses and all critiques, then writes the final answer. It combines the best insights, resolves disagreements, and produces a balanced, comprehensive response.
Tested March 2026 against Perplexity, GPT-4o-mini, Claude, and Gemini across 4 complex questions. Scored on balance, structure, depth, nuance, and actionability (50 points total). Deliberation produces higher quality than search or single-model responses.
Methodology: Heuristic scoring across 4 complex questions (March 2026). Models: Perplexity sonar-pro, GPT-4o-mini, Claude 3.5 Haiku, Gemini 2.0 Flash, Grok-3. Perplexity comparison: 36% higher quality, 4/4 wins. Full details in BENCHMARK_RESULTS.md.
Every feature is designed to catch the mistakes that single AI models confidently make. Enable any combination for your question.
3+ models from different providers (OpenAI, Anthropic, Google) answer independently, then critique each other. Catches blind spots any single model would miss.
Extracts every factual claim from the final answer, checks each against sources using Chain-of-Verification (CoVe), and shows a confidence score. You see exactly which claims are verified.
One council member is assigned to actively challenge the group consensus. Prevents groupthink and ensures controversial topics get both sides represented.
Compares model responses semantically. If they fundamentally disagree, you get a warning that the topic is contested — not a false sense of certainty.
Based on Stanford's STORM methodology. Generates multi-perspective research questions, conducts expert interviews across models, and produces a structured report with executive summary, findings, and limitations.
Adds numbered [1][2][3] references throughout the answer, with a source list showing exactly where each claim comes from. Know which sources support which statements.
Multiple ways to interact with the council, designed for different workflows and devices.
Advanced multi-agent UI with real-time streaming, debate visualization, and structured synthesis. The premier experience.
/v2Browse, search, and export your past council sessions. Filter by mode, date, or topic. View full deliberation logs and re-run sessions.
/history