About AI Board
One question, several AI models, one synthesized best answer.
What it does
The AI Boardorchestrates a structured discussion between multiple AI models — OpenAI GPT, Anthropic Claude, Google Gemini, anything on OpenRouter, or your own OpenAI-compatible endpoints such as Ollama and LM Studio. You ask one question; the models answer, read each other's responses, and critique and refine their positions over several rounds. When the discussion converges (or the configured effort level is reached), a judge model you choose reads the whole transcript and writes a single, synthesized final answer with a confidence score and any notable dissent.
One strong model can be wrong with great confidence. Several models that must defend their answers against each other catch more mistakes, surface more perspectives, and produce a better-grounded result than any single response.
Discussion modes
- Collaborative panel— models work together, building on each other's answers toward consensus.
- Debate — models take opposing positions and argue them, useful for decisions and trade-offs.
- Specialist + reviewers — one model drafts the answer while the others review and demand fixes.
- Build mode — an architect model breaks a coding task into work items and worker models implement them in parallel waves. Files can be written to a real project folder on your machine, downloaded as a zip, or applied through an optional local runner that adds shell access, MCP tools, and a browser control panel to pick the project folder, watch live activity, manage MCP servers, and keep itself current with a signed self-update.
Local-first and private by design
There is no backend. The app is a static site; the orchestration engine runs entirely in your browser tab and calls the AI providers directly with the API keys you enter on the Settings page. Keys, discussions, and attachments are stored in your browser (IndexedDB) or in a local folder you pick — optionally encrypted with a passphrase using AES-256-GCM. Nothing is ever uploaded to a server of ours, because there isn't one.
Use an AI account you already pay for
If you subscribe to ChatGPT Plus/Pro or GitHub Copilot, you can point AI Board at that account instead of entering per-token API keys. A small account runner runs on your machine, you authorize it once with OpenAI or GitHub, and the app talks to it over localhost — the login stays in a file on your computer.
A connected account works everywhere the app uses models: discussions, Build mode, and the games. That last one is the fun part — you can play chess, Connect Four, Battleship, Codenames, or Fireworks against GPT-5.5, or sit out and watch two models play each other.
This draws on your account's own entitlement, so rate limits, model availability, and behavior are set by OpenAI or GitHub rather than by AI Board.
Contact
For feedback, bug reports, privacy questions, or collaboration, email [email protected].
Source code is available on GitHub.
Learn more
These short guides explain the main workflows without adding more navigation to the app itself.
FAQ
- Is it free?
- The app itself is free and open source. You pay your AI providers directly for the tokens the discussion uses; the app shows a cost estimate before you start.
- Which API keys do I need?
- At least one provider key (OpenAI, Anthropic, Google, or OpenRouter) — two or more models make discussions worthwhile. Local models via Ollama or LM Studio need no key at all.
- Can I use it offline?
- The app loads as a static site, but discussions need network access to reach the AI providers — unless you use only local models on your own machine.
- Can I use my ChatGPT subscription instead of an API key?
- Yes. Connect a ChatGPT Plus/Pro or GitHub Copilot account through the local account runner and its models appear alongside your API-key models — in discussions, in Build mode, and in the games. Rate limits and availability are governed by your provider account.