Wollipog

Wollipog

Wollipog is a local-first control plane for running and supervising coding agents across development machines. It gives you one browser or desktop interface for sessions, approvals, worktrees, terminals, diffs, reviews, automations, and remote runners while execution stays on the machine that owns each repository and toolchain.

[!IMPORTANT] Wollipog is under active development. Release binaries are currently unsigned, and remote access should be limited to trusted tailnet devices.

Architecture

Browser or Desktop UI
  -> Control Plane API       HTTP commands and WebSocket events
    -> Runner                outbound authenticated WebSocket
      -> Agent Driver        ACP stdio, Claude Code, or Codex
        -> Local Repository  shell, Git, tools, and isolated worktree

The runner owns provider processes and filesystem access. The control plane stores normalized session state and events, while the UI remains a control surface. ACP stays runner-local and is never exposed directly to the browser.

See [Concepts and Glossary](docs/concepts-and-glossary.md) for the relationship between Instances, Machines, Runners, Projects, Locations, Workspaces, and Sessions.

Current Capabilities

• Connect native, WSL, SSH, container, and operator-configured cloud execution targets.
• Discover supported coding-agent installations and their authentication state.
• Run Claude Code and Codex through native drivers, plus compatible agents through ACP.
• Create isolated Git worktrees and inspect status, diffs, commits, branches, and pull requests.
• Stream agent messages, reasoning summaries, plans, tool calls, file changes, terminal output, and errors.
• Handle approvals, structured questions, authentication prompts, and policy decisions.
• Fork or resume supported conversations while preserving filesystem provenance.
• Coordinate multi-agent runs, pods, workflow