AgentDock is a tool I built for working with several coding agents without losing track of what each one is doing. A mission starts with a plan I can review and edit. Once I approve it, the agents work on bounded tasks in separate Git worktrees, with their progress and changes visible in one interface.
I can pause an agent, send it a message, inspect its changes or open its terminal. The mission keeps its conversation and execution state, so there is a way to pick up interrupted work. It uses the existing Codex CLI login, and a preview mode lets me try the planning and control flow without running real agents.
The planner can also decide that a request only needs an answer or is already satisfied, so it does not have to create workers for every mission. For work that does need execution, I wanted to see the task boundaries and dependencies before starting it. That decision is reflected in the plan review and the controls available while agents are running.
I wrote it in Python and keep the source on GitHub. The screenshot shows the application with a fresh workspace, before any missions have been started.
