langchain-ai/openwork ? reverse-engineered prompt

Build me a desktop app called openwork that lets me chat with an AI agent for coding and file work on my computer.

I want it to feel like a simple local workspace where the agent can read and edit files, plan out tasks, and hand off parts of the job to subagents when needed. It should also be able to run shell commands, but always make me review and approve anything risky before it happens.

Please make the app easy to start with npx or a global install, and also runnable from source in dev mode. Include a settings panel where I can pick or configure supported AI models from Anthropic, OpenAI, and Google.

Keep the UI clean and desktop friendly, with a place to see the agent conversation and what actions it wants to take. If you need to look up current docs online, go ahead.