I created a containerized VS Code server with the Goose AI coding agent integrated, and recently added two features I thought might be useful:
- A shared tmux terminal so multiple users can watch and interact with the same Goose AI session simultaneously.
- A REST API that lets other systems send commands to the terminal and retrieve conversation logs programmatically.
The main goal was to create a coding environment that could be used both interactively by humans and programmatically by other LLM systems. You can run it as a collaborative tool where several people work with the same AI agent, or you can control it entirely through API calls.
The setup uses Docker, so you can easily spin up clean environments for specific projects. You can configure Git credentials during runtime, which makes it suitable for automated workflows that need to clone repos, make changes, and create PRs.
I'm sharing this because it seems like a practical approach to building LLM-powered developer tools that can be both interactive and programmable.
