Documentation
Understand your workspace.
Look up how Prokop works, from workspace behavior to configuration and commands. For a first session, follow Get started. For a specific task, use the how-to guides.
Start here
Core concepts
Reference
Common questions
What is Prokop?
An open-source coding workspace for Claude Code, Codex, and its own agent runtime. One binary runs a local daemon with the full web client inside: sessions, worktrees, files, Git, terminals, approvals, memory, learning, and scheduled work. You drive it from a browser, desktop or phone.
Where does everything live?
~/.prokopai/- server data, database, agents, models<project>/.prokopai/- workspace memory and MCP configuration<project>/AGENTS.md- instructions the agent loads for that project<project>/.agents/skills/- workspace skills
Your project context and local database stay on disk under paths you control.
Do I need to edit config files?
No. Providers and models connect in the client under Settings → Providers & Models, and Claude Code or Codex need only their own login. API keys, subscriptions, model selection, memory and learning: all UI. API keys can also live in .env, alongside server settings such as the auth token or TLS.
Can I use it from my phone?
Yes. The client is mobile optimized and served by your daemon: same sessions, same agents, same permissions, reorganized for one surface at a time. Turn on remote access, pair the phone with a four-digit code, and reach it over networking you choose such as Tailscale. Installing the PWA on a phone needs HTTPS, which Tailscale HTTPS or the built-in TLS listener can provide.
What about security?
By default the daemon listens only on your machine. Other devices reach it after you turn on remote access and pair them; each can be removed at any time. Separately, commands and file operations that reach outside the workspace, touch secrets, or are destructive ask for approval, with the risky part highlighted.