Inspiration
Parallel coding agents make implementation faster, but they also make coordination harder. When several agents work in separate Git worktrees, it is easy to lose track of which branch owns a change, where files overlap, and which terminal needs attention.
What it does
Fleet Radar is DevIsland’s local-first control tower for parallel coding agents. It discovers active worktrees and live agent sessions on the developer’s Mac, then shows branch state, modified files, overlap warnings, and priority work in one Session Center view.
Developers can spot shared file paths before they become merge conflicts and jump directly from a Fleet card to the relevant terminal. The feature keeps repository metadata local—no source code or worktree data is sent to a cloud service.
How we built it
Fleet Radar is implemented in Swift and SwiftUI inside DevIsland, a macOS menu-bar and notch-overlay app for coding-agent activity. We modeled each worktree as a local session context, derived overlap signals from modified paths, and connected card actions to terminal focus. The demo uses three concurrent worktrees: two intentionally modify the same file while a third works independently.
Built with Codex and GPT-5.6
I used Codex and GPT-5.6 throughout the project: to explore the existing session architecture, design the Fleet Radar data flow and SwiftUI surfaces, implement the feature, and verify the multi-worktree scenarios. Codex also helped turn the working demo into a concise submission video.
Challenges and lessons
The hardest part was giving developers useful coordination signals without overstating them. A shared path is an early warning, not necessarily a merge conflict, so Fleet Radar makes that distinction explicit. Keeping scans local and lightweight was equally important because this information needs to be useful while agents are actively running.
What's next
Next, Fleet Radar can grow into richer team coordination while preserving its local-first design: clearer work ownership, more conflict context, and better handoff views for parallel agents.
Log in or sign up for Devpost to join the conversation.