mukarram-one/google_adk ? reverse-engineered prompt

Build me a local multi agent assistant that runs with Google ADK and Gemini, like the README describes. I want one main coordinator agent that can hand tasks off to a few specialists, one for reading text from images or PDFs with OCR, one for browser actions like checking open tabs and searching the web or YouTube, one for research and reading URLs, one for drafting clear content or code, and one for reviewing and polishing the final result.

Make it easy to run locally in a browser, with a simple web UI for chatting with the agents. It should read settings from an env file, support my Gemini key, and work with Tesseract if I have it installed. If anything is unclear, look up the current Google ADK docs online and wire it up in the cleanest way. I want the project to feel like a real working demo I can try right away and learn from.