AI Board
Multi-model AI discussions that synthesize the best answer.
New Discussion
Ask a question and let multiple AI models discuss, debate, or build together before delivering the best result.
Topic
e.g. Compare Postgres vs. MongoDB for a booking app, or build a recipe manager web app
Images, documents (PDF, Word, text), audio, and video. Models incompatible with attached types will be disabled.
How Collaborative Panel works
Each model gives an initial answer, then critiques and improves the group output across later rounds.
Clear and reasonably thorough. The default.
Added to every model's instructions for this run.
Each model's built-in setting (OpenAI medium, Claude high).
Add API keys and enable providers in Settings.
Recent Discussions
Click to view live or completed results
No discussions yet.
One question, several AI models, one answer
AI Board runs a structured discussion between models from OpenAI, Anthropic, Google, OpenRouter, and any OpenAI-compatible endpoint such as Ollama or LM Studio. Each model answers your question, reads what the others said, and refines or challenges its position across several rounds. A judge model you choose then reads the whole transcript and writes one synthesized answer with a confidence score and any notable dissent.
A single model can be wrong with great confidence. Several models that have to defend their reasoning against each other catch more mistakes and surface more perspectives than any one response.
Ways to run a discussion
- Multi-model discussions — models collaborate toward a consensus answer.
- AI debate tool — models argue opposing positions to expose trade-offs.
- Build mode — an architect model plans coding tasks and worker models implement them in parallel.
- Games — play chess, Connect Four, Battleship, Codenames, or Fireworks against the models.
Local-first by design
There is no backend. The orchestration engine runs entirely in your browser tab and calls the providers directly with the API keys you enter on the Settings page. Keys, discussions, and attachments stay in your browser or in a local folder you pick, optionally encrypted with a passphrase. Nothing is uploaded to a server of ours, because there isn't one.
Read more about how it works, or set up the Runner V2 agent kernel that Build mode uses for real file and shell access.