Two AI debaters argue any motion while a judge keeps score.
Pick a motion and a style. The proposition and opposition take turns through openings, rebuttals and closings, then an adjudicator scores logic, evidence and rebuttal. You can also argue one side yourself.
The app itself, running inside this page. Anything you do here is saved in your browser.
Ask it to do things
Press Ctrl+K in the app (Cmd+K on a Mac) and describe what you want. The model chooses from the app's own functions, runs them in order and tells you what changed. Some examples:
| Debate whether cities should ban private cars, three rounds | Sets the motion and round count, then starts the debate. |
| Make the opposition argue like a skeptical economist | Rewrites the opposing persona before the next round. |
| Who is winning so far and why? | Reads the adjudicator's scores and summarizes the clash. |
Instructions you save in the same box, such as "keep it short" or "I'm a beginner", are added to every AI request the app makes.
Features
- Configurable personas
- Style (formal / spicy / Socratic / ELI12) and number of rebuttal rounds.
- Live streaming speeches
- With full-transcript context for real rebuttals.
- AI judge returns structured JSON
- Rubric scores, clash map, margin, advice.
- SVG radar chart comparing both sides
- On logic, evidence, rebuttal and delivery.
- Local analytics
- Flesch readability, evidence markers, hedge words, direct call-outs.
- Export the full debate
- And verdict as Markdown.
- You-vs-AI mode
- Argue either side yourself with a 2-minute speech timer while the AI rebuts your actual points.
- Oxford-style audience vote
- Record your opinion before and after; the side that moves the most minds wins the room.
- AI fact-checker that extracts factual claims
- Rates each one and computes a credibility score per side.
- Debate library
- Every finished debate is saved locally and can be reopened.
Runs in your browser
Readability scoring, syllable counting, rhetoric metrics, radar chart rendering, export. None of this needs a network connection or a model.
Uses a language model for
One streamed chat completion per speech, each with its own role-play system prompt. A low-temperature, JSON-mode adjudicator call that scores the transcript. Claim extraction + verification pass feeding a local credibility score.
Notes
- Cost
- Free. Each account has a daily allowance of AI requests.
- Your data
- What you make stays in this browser on this device. Your account is only used for signing in and for the built-in AI. You can export a backup or clear everything from Settings.
- AI access
- Signed-in users get the built-in AI with nothing to set up. You can also use your own Groq or OpenRouter key under Settings.
- Source
- The code and tests are on GitHub under the MIT license. The tests also run in your browser.