Spotlight: ADE — a worktree-first cockpit for your coding agents
Most AI coding tools want to be the agent. ADE isn't trying to be one. It's the layer that sits above Claude Code, Codex, Cursor, OpenCode, and Droid, and lets you run several of them at once without losing track of what's happening where.
The problem it's aiming at is a real one for anyone using AI agents seriously: once you're running more than one task at a time, you end up with a pile of terminal tabs, branches you forgot you made, and no good way to compare what two different agents did with the same prompt. ADE's answer is to make the git worktree the basic unit of work. Describe a task, and it spins up an isolated branch, worktree, checkout, terminal, and port for it automatically. Those worktrees stack on a graph so you can see dependencies and rebase order at a glance instead of guessing which branch depends on which.
The genuinely smart part of the build is the sync model. The same worktree and chat list shows up in a terminal-native mode, the desktop app, and an iOS companion, and they stay mirrored, messages, tool calls, and proof captures included. Start a task on your laptop, check on it from your phone, finish it back at your desk. That's a small thing to say and a hard thing to build correctly, and it's the reason ADE reads less like an IDE plugin and more like infrastructure for running agents at scale.
A few other details show real product thinking rather than feature-padding. Grid view tiles every parallel run side by side so you can watch multiple agents work at once. Subagents let you hand a lead agent a plan and have it dispatch and track a fleet of workers. PRs get generated straight from a finished chat, diff and checks included, without a trip to GitHub. There's a built-in browser pane for web QA and a Linear integration so issues can move through states without leaving the app. None of this replaces the agents themselves, it just gives them a place to run where you can actually see and manage them.
Who should try it: developers who are already running more than one AI coding agent, or running the same agent on multiple tasks at once, and who are comfortable with git worktrees as a working model. If you're doing serious parallel AI-assisted development, on a Mac or Windows machine (or Linux via the terminal build), and you've felt the pain of losing track of five different agent sessions, this is built for exactly that pain.
Who shouldn't bother: if you use one AI coding assistant for one task at a time, ADE is overhead you don't need, a single Cursor or Claude Code window will serve you fine. It also assumes you're already using one of the supported agents, it doesn't ship its own model or agent, so if none of Claude Code, Codex, Cursor, OpenCode, or Droid are in your stack, there's nothing here to orchestrate yet. And the "free forever" line in the tagline isn't explained anywhere on the site in terms of what happens with remote runtimes or team usage down the line, worth watching before you build a workflow around it.
Where I'd guess this goes next: the Linear integration and remote runtime support both point toward team use, not just solo indie hacker workflows, someone running a small eng team with multiple people dispatching agents against the same repo. That's a natural expansion, but it's also where the sync-everywhere architecture gets harder to keep simple. Worth watching whether ADE stays a clean local-first tool or grows into something that needs a hosted backend to make the team story work. That's a read, not something the site claims.
Try ADE: ade-app.dev
See the launch: ADE on welaunch.sh