yuzc-001/ai-roundtable-room

AI Roundtable Room is a tool that takes a complex question and runs it through several AI roles acting as reviewers, then produces a structured decision package you can save, follow up on, or paste into a project. The README, which is written in Chinese, frames the idea as an alternative to asking one model and getting one tidy answer. Instead, the system has a host, a strong-opinion voice, a risk breakdown role, an evidence checker, and a user perspective role that take turns. The output keeps track of disagreements, evidence status, open questions, risks, and action items. The authors suggest it for product, technology, growth, and business judgement calls where you want to see alternative plans, objections, risks, and conditions for launch. They also list writing, legal, consulting, and retrospective work, plus long running project discussions where you want to keep confirmed conclusions and reopen them later. The README says API keys stay on the server side and are not exposed to the browser. To run it locally, Windows users run start.bat and macOS or Linux users run bash start.sh. The first run copies .env.example to .env, and at minimum you fill in an OpenAI compatible API key and model name. The app then opens at http://127.0.0.1:5173. You can also configure multiple providers such as DeepSeek, GLM, or Claude, and pin specific roles to specific providers. Common npm scripts include dev, doctor for checking the env, test using Vitest, build, start, and a check script that runs tests plus a dependency audit. There are also video scripts that use Remotion to render an explainer video of the project. For public deployment the README recommends setting an access code, a session secret, and a daily meeting limit so other people do not burn through your model quota. A Dockerfile is included. The project layout has an Express server, a React frontend in src, shared persona definitions, a Remotion video folder, and docs covering architecture, a security audit, and a roadmap. License is MIT.