for cli agent users
Keep the agent. Lose the terminal farm.
You already run a coding agent in a terminal and it codes well. This page is about everything around the agent, because that is what daily agentic work actually spends time on.
What stays the same
- Your models. Bring API keys or Codex subscription authentication. Prokop does not lock you to a provider.
- The coding ability. Prokop is not a different agent. It is the environment one works in.
- A real terminal. Persistent terminals live inside the workspace, so long tasks keep running while you keep talking.
What changes
- Six sessions stop meaning six windows. The board keeps six live panes in view. Any number of sessions can be running; the server is not limited to six.
- Project switching stops being context juggling. Workspaces hold their own sessions, memory, and tools. An Overview aggregates sessions across projects.
- Process sprawl ends. One daemon on your machine. Close the terminal or the browser; sessions and terminals keep running.
- You get real surfaces. Files, diffs, worktrees, commit history, and permissions in the client, not just stdout scrolling by.
- Context survives. Agent memory follows the agent across projects, workspace memory stays with the project, and full-text session search brings old decisions back. All of it is opt-in, plain files on disk.
What it costs you
An interface sits between you and the agent. If you want zero chrome and a single REPL, Prokop is more, not less. And it is a young project maintained by one person: evolving, not hardened. Apache 2.0, so the code outlives any roadmap either way.
Try the workflow
One install command and prokop init. Point the first workspace at a project you already know and run two sessions side by side.