The mission-control model
Nexus treats AI agents as a fleet you dispatch and review — not tools you drive keystroke by keystroke. The whole design follows from that shift.
The shift
AI coding agents changed the shape of the work. They are:
- Plural — you run several at once, across several repos.
- Asynchronous — you dispatch and come back, you don’t babysit.
- Long-running — a single task can churn for minutes.
A row of terminal windows is the wrong interface for that. Mission control is the right one.
The project as the unit
In Nexus, the project is the primary unit of navigation. Selecting a project performs a one-click rebind of the entire app: the rail, branches, worktrees, sessions, and review all snap to that repo. You are never reconstructing context across windows.
Two modes
| Mode | Purpose |
|---|---|
| Workbench | Dispatch runs, watch live sessions, manage projects, branches, and worktrees. |
| Review | Walk an agent’s diff hunk by hunk — approve, request changes, or skip. |
Principles
| Principle | What it means |
|---|---|
| Dispatch, don’t drive | You describe intent; the agent executes. Nexus optimizes the dispatch-and-review loop, not keystroke entry. |
| Provider choice per run | Pick the agent per task. The same project can run Claude Code on one task and Copilot on the next. |
| Isolated by worktree | Every run gets its own worktree, so concurrent agents never trample one working directory. |
Nexus is deliberately not an editor. It does not compete with VS Code or Rider — it sits beside them and owns the parts they handle badly: fleet status, dispatch, and cross-repo review.