Getting Started

1. Select two or more AI models from the Models panel.
2. Choose your mode: Standard (with optional Debate/Critique toggle), Steelman, Crucible, Go/No-Go, Duel, or Gauntlet.
3. Optionally upload a document or add Pinned Topics to focus the debate.
4. Enter your prompt and click BRIDGE IT.
5. Click Next Round after each round. Run at least 2 rounds to unlock Sync Scoring and 3 rounds to render a verdict.

Understanding the Panels

• Revised VersionThe latest jointly refined answer after each round of critique.
• CritiqueThe synthesized critique from all non-drafter models. When three or more models are selected, all critics contribute independently and their critiques are merged into one.
• ChangelogA concise list of what changed between the current and previous revision.
• Sync EstimateA multi-dimensional score showing how closely the models agree, broken down by factual agreement, framing alignment, scope agreement, ethical alignment, and confidence level. Unlocks after Round 2.
• Dissenting ViewsImpasses and minority opinions preserved even when they did not make it into the consensus revision.
• Diff ViewA word-level comparison highlighting exactly what changed between revisions.
• Steelmanned VersionIn Steelman mode, the Round 0 synthesis showing the strongest possible version of your argument before critique begins. Includes a Strength Score out of 100.

Debate Modes

• 🗣 StandardAll selected models critique each draft independently. Critiques are synthesized into a unified review. Use the Debate/Critique toggle to choose between collaborative synthesis (Debate) or adversarial evaluation (Critique).
• 💪 SteelmanModels first strengthen your argument in Round 0, then debate the reinforced version. Produces stronger, more defensible consensus. Check 'Strengthen only' to skip the debate rounds.
• ⚔️ DuelTwo sides submit opposing arguments independently — Side B's submission is hidden until after they respond, preventing anchoring bias. Bridget synthesizes both positions across multiple rounds, producing steelmans, critiques, common ground, genuine dispute identification, and a final synthesis. Watch live duels in the Lobby.
• 🔥 CrucibleSix targeted attack rounds: Financial, Technical, Reputational, Legal, Operational, and Edge Cases. Use when you want your argument stress-tested from every angle before committing.
• 🚦 Go/No-GoSubmit a proposed decision. Bridget rules GO, NO-GO, or GO-WITH-CONDITIONS with explicit reasoning. Forces 6 rounds. Use when you need a clear call, not analysis.
• Workflows
• ⚡ GauntletA four-stage pipeline that automatically chains Steelman, Crucible, and Go/No-Go into a single end-to-end analysis: Frame → Strengthen → Attack → Rule. Use when you want the full treatment in one button press.

Features

• 😈 Devil's AdvocateAssigns one model per round to challenge the emerging consensus, preventing groupthink. Use the three-stop slider to control behavior: Off disables it; Constrained (yellow 😈) grounds all challenges strictly in the submitted text; Full Aggression (red 😈) gives the DA free rein to speculate beyond the document.
• ✨ Magic — Prompt AssistantAnalyzes your prompt and suggests a refined version along with recommended session settings — mode, intensity, Devil's Advocate state, and round count. A confirmation window lets you review and edit all suggestions before anything runs.
• ⚡ AutopilotUses Magic to suggest settings, then starts Speed Round automatically without intermediate confirmation.
• 🔮 Beit DinModels take turns presiding in three-round chunks; the presiding model sits out the debate and writes a checkpoint ruling.
• ⚔️ LobbyBrowse and watch live duels between two sides in the Lobby.
• Daily limitsGuests get 9 rounds a day; sign in for more.
• 🔄 Multi-Model SynthesisWhen three or more models are selected, all non-drafter models critique independently in parallel. A dedicated synthesis call merges the strongest points from every critic, resolves contradictions, and preserves genuine cross-model disagreements as impasses. More models means deeper critique, not redundancy.
• Follow UpAdds a user note between rounds that all models will see and incorporate in the next revision.
• Pinned TopicsLocks specific sub-questions that every model must address explicitly in each round.
• Document UploadPaste or upload a file; its text is included as shared context in every round.
• TemplatesPre-built prompts organized by category for common tasks including legal, business, research, finance, and more.
• 📋 VerdictRender a session as a permanent, shareable artifact. Includes the final consensus, sync score, dissenting views, and pinned topic resolutions.
• 🔀 CompareCompare two existing verdicts side by side, with a Divergence Summary showing where their conclusions align or differ.
• CrownsCrowns are earned by sharing a rendered verdict to verdicts.ai. Crowns are game points with no cash value. They cannot be bought, sold, redeemed, or exchanged for money, Bitcoin or any other currency.
• Donate to the MuseumThe ₿ and 👑 buttons open the donation page: Bitcoin for anyone, Crowns for signed-in users.
• ⚡ Speed RoundRuns rounds automatically and stops when the models converge (95% sync for two rounds by default, adjustable) or when revisions stop changing. Checkpoints at rounds 3, 6 and 9.
• 🔒 Secure Key StoragePro, Collaborator and Inner Circle accounts can store API keys server-side using AES-256-GCM encryption. Keys survive browser clears and device switches and never leave the server unencrypted.
• 🔀 Debate/Critique ToggleStandard mode only. Switch between Debate mode (collaborative synthesis, models build toward consensus) and Critique mode (adversarial evaluation, critic leads with problems, no revision of the original).
• ⚡ Intensity SliderControls how the critic delivers feedback — Supportive (diplomatic), Balanced (balanced), Rigorous (direct and demanding).
• 📝 Transcript Analysis TemplateUpload a legal transcript or hearing record alongside a protocol or checklist. Bridget audits execution against the standard with page and line citations and downstream consequence analysis.
• ✦ Verified VerdictRender a permanent credentialed artifact with a unique Verdict ID, UTC timestamp, model roster, and sync score. Shareable via permanent link. Verifiable by anyone with the link.

Legal and Professional Use

Bridget works best for stress-testing arguments, not verifying facts. For legal work: conduct your own case law research and citation verification first, then upload verified materials as document context. Bridget will stress-test your argument against those materials through adversarial multi-model debate. Bridget is the crucible — not the library.

Bridget supports 11 citation sources that auto-inject source text when a reference is detected in your prompt: Legal: CourtListener, Govinfo (US Code + CFR), Congress.gov, Federal Register. Science & Academic: PubMed, arXiv, Google Scholar. Scripture: Sefaria, Bible (KJV), Quran. Financial: SEC EDGAR. Some sources return abstracts or summaries rather than full text.

Enable sources in the ⚖️ Legal Research, 🔬 Science & Academic, and 📖 Scripture panels, then include a reference ID (e.g. PMID, arXiv ID, FR document number) or an explicit Scholar: query in your prompt.

Privacy

When you run a session, Bridget saves your prompt, any document text you submit, and each round's responses, so your session history and verdicts can be kept. A rendered verdict keeps its prompt and verdict data; a verdict you share to verdicts.ai is public. Your prompts are sent to the AI companies whose models you select (such as OpenAI, Anthropic, Google, xAI, Venice.ai, DeepSeek, Mistral, Cohere or Meta (Llama)), using your API keys or Bridget's system keys; each company's own data policy applies. API keys you enter are kept in your browser (for the current tab for guests; on your device for signed-in users). Eligible signed-in accounts can instead store keys on Bridget's servers, encrypted with AES-256-GCM. Bridget's operational logs record counts and status, not the content of your prompts or answers.

Tips

Mission

Bridget was built to bridge AI platforms — to make them argue, synthesize, and converge. It became something more: accountability infrastructure for a world that needs verified truth.

The ecosystem has three layers. Bridget is where models debate. verdicts.ai is the public record — where each verdict lives permanently, verifiable by anyone. Common Knowledge is the synthesis layer — where shared verdicts accumulate into searchable, citable ground truth. Every language has a word for truth. Now they all have an address.

3% of revenues are committed to The Bitcoin Museum Trust — a permanent institution being established in Brooklyn, New York. Both technologies decentralize truth. Bridget debates it. Bitcoin records it.

Bridget is a proud sponsor of The Bitcoin Museum.