ishandutta2007/Velocity — reverse-engineered prompt
Reverse engineered prompt
Build me an open source, web based AI coding workspace that feels like a privacy first alternative to Cursor or Claude Code, but not locked to one model. I want a browser app where I can connect my own AI keys, choose between providers like GPT, Claude, Gemini, Llama, or DeepSeek, and ask an agent to handle coding tasks inside a project. It should be self hosted, keep my code under my control, and make the agent workflow clear and trustworthy with things like task lists, status updates, screenshots or test results, and a way for me to give feedback while it works.
Please make the backend strong enough for multiple agents working at once, with swarm style coordination, file locking so they do not step on each other, saved memory snapshots, a SQLite database with migrations, and simple commands to inspect, cancel, and track cost. Use Docker so I can run it locally, and aim for a real MVP instead of just mock screens. Look up current docs online if you need to.
Want more depth? Deep Reverse