This build is a foundation rebuild of the Carrot chat + StudyVault study workspace.
- Copy
.env.exampleto.env. - Put your OpenAI API key in
OPENAI_API_KEYif you want cloud chat, vision, web search, image generation/editing, transcription, and speech. - Run
START-CARROT-AI.baton Windows, or runpython server.pyand a static web server from this folder. - Open the StudyVault page.
The API key stays on the server and is never placed in browser JavaScript.
Supported browser-first imports include PDF, DOCX, legacy DOC (server fallback), TXT, Markdown, CSV, JSON, HTML, RTF, PNG/JPEG/WebP/BMP/GIF/SVG, and folders.
PDF.js, Tesseract, and JSZip are bundled locally. DOCX extraction no longer depends on a CDN. The local Carrot server also provides OCR and legacy Word/PDF extraction fallbacks.
Cloud AI routes through the local Carrot server. The default text model is gpt-5.6-sol and the default image model is gpt-image-2.
Image requests use the real image-generation API path when cloud AI is configured. Editing and variations send the previous image back to the image pipeline rather than drawing a fake canvas result.
Without a cloud API key, the app remains usable in local/offline mode and clearly labels fallback behavior.