Yash3510/Multi-LLM-Agentic-Ai ? reverse-engineered prompt

Build me a local first AI workbench that runs completely on my own machine and never sends data to the cloud.

I want it to take one request, figure out what kind of task it is, route it to the right local model, pull in relevant files from my own documents, generate an answer or deliverable, then have a second model check the result before it is shown. If the check fails, it should try again once. After that, a person should still have to approve it before anything is released.

Please make the whole flow visible in the app, with live status updates and a clear breakdown of what each step did and why. I also want support for running generated Python safely in a locked down sandbox, and for exporting approved results as a formatted Word file, spreadsheet, or slide deck when needed.

Use current docs online if you need to, but keep everything working with a local model server only.