What is adk-okf-grounded-chat?

This project is a chat application built to explore a different approach to giving AI agents reliable knowledge, instead of the usual method where a system searches through documents using a vector database in the background. It is built on Google's Agent Development Kit and something called the Open Knowledge Format, which stores an agent's knowledge as plain Markdown files tracked in git, so changes go through normal version control instead of being hidden inside a database. The idea is that the AI agent checks its local, curated knowledge files first before falling back to its own general reasoning. You can drag and drop files like PDFs, images, audio recordings, or paste in a URL, and the app will convert them into these Markdown knowledge files on the fly during your session. There is also a fully offline mode called Pure OKF Mode that answers questions using local rule-based text matching instead of calling any AI model at all, keeping everything private. The app also lets you edit, add, or delete these knowledge files directly through its interface, and it includes a fallback responder for common questions that keeps working even if the connection to the AI service is down. When the AI does answer, its reasoning steps are shown in a collapsible section in the chat so you can see how it arrived at an answer. The project has two parts: a React and Vite frontend for the chat interface, and a Python FastAPI backend that runs the agent logic and streams responses back to the browser. To run it, you need Node.js 18 or newer and Python 3.10 or newer, install dependencies for both the frontend and backend separately, and then run a single startup script that launches both parts together.