Chat with RTX, minus the RTX requirement

What it does

Dot is an Electron-based desktop app that loads folders of documents—PDF, DOCX, PPTX, XLSX, and others—into a fully local RAG pipeline built on FAISS, LangChain, and llama.cpp. It defaults to Microsoft’s Phi-3.5 model and offers a general-purpose “Big Dot” chat mode for questions unrelated to your files. Everything runs offline, and the project explicitly targets users without a programming background.

The interesting bit

The app is essentially a curated glue layer around battle-tested local-AI tools, inspired by Nvidia’s Chat with RTX but decoupled from green-team hardware. It is also a student side project, which explains both the ambition and the visible rough edges.

Key highlights

• Ships as a standalone download for Apple Silicon and Windows; Linux is still on the wishlist.
• Uses FAISS for local vector storage and Hugging Face integrations for model handling.
• Defaults to Phi-3.5, with the developer noting that LLM choice is now configurable.
• Supports general chat alongside document-grounded RAG via a dual-mode interface.
• Repository description advertises all-local TTS and voice features (tts, whisper-cpp topics), though the README focuses on chat and does not detail how voice features are exposed.

Caveats

• Linux is not yet supported; the download page only lists Apple Silicon and Windows.
• The README notes that the developer is a student managing this on the side, so pace and polish may vary.
• File loading currently requires selecting folders rather than individual files, a known friction point.

Verdict

If you want a no-cloud, no-subscription document assistant and are comfortable with Electron’s resource appetite, Dot is worth a look. Linux users and those needing mature production polish should wait or contribute.