simpleyyt/ai-manus ? reverse-engineered prompt

Build me a web app for an AI agent that can actually do tasks for me, not just chat.

I want a frontend where I can start a conversation, watch the agent think and act in real time, and switch between different sessions. The agent should be able to use a browser, run terminal commands, work with files, search the web, and connect to extra tools if needed. Each task should run in its own sandbox so it feels safe and isolated.

Please also include user login, file upload and download, the ability to stop or interrupt the agent, and a library area where I can browse saved files and artifacts from previous sessions, search them, favorite them, and jump back to where they came from.

Make it work in Docker with a simple setup, and keep it compatible with common chat model APIs that support tool calling. If you need current docs for anything, look them up online.