What is codevinci?

CodeVinci is a locally run tool that turns design mockups into working webpage code. You draw or import a design, and the tool calls a vision capable AI model to generate the corresponding HTML. The result appears in a live preview panel on the same screen, isolated inside an iframe so it does not interfere with the tool's own interface. The tool has two main modes. Canvas mode lets you draw on a canvas, manage layers, and trigger the AI to generate or update HTML from what is on screen. Voice mode lets you describe changes by speaking out loud, and the tool applies those changes as a stream of updates in real time. The README notes that voice mode is now well developed, while canvas mode is still in progress. Alongside these, there is a source code editing view where you can manually edit the generated HTML, and open or save files directly. Running CodeVinci requires a working Node.js environment and an API key for a vision capable AI model, either from OpenAI or Anthropic, depending on which API format you configure. Voice mode requires a separate Deepgram API key for speech recognition. Setup involves cloning the repository, running npm install, copying a sample .env configuration file, and filling in the API keys and model name before starting the local development server. By default the tool runs at a local address in your browser, and separate npm commands exist for building the frontend and backend, running the production build, and running a TypeScript type check. The tool does not send your project files to any outside service beyond the AI API calls you configure yourself. Full configuration and usage details are documented in a separate USAGE file in the repository. The project is released under the MIT license.