You're the whole team now — code, product, QA, all of it. dev-3.0 is the cockpit for commanding a fleet of AI coding agents at full speed, while one board keeps you focused. Without losing your mind.
Your projects, your boards, your agents — your flow.
You're running 5+ agents across different terminals, repos, and branches. Switching context takes forever. You lose track of what’s where. We tried every tool in the space — each had cool ideas, but something was always missing. So we built our own. The hard part was never typing code — it’s keeping your focus across all of it.
Dozens of terminal tabs, each running a different agent in a different branch. Finding the right one is a nightmare.
Every time you switch between agents, you waste minutes remembering where you left off and what each one is doing.
Multiple agents editing the same repo. Constant merge conflicts. No isolation between parallel workstreams.
Projects, tasks, and terminals — each layer gives you more control over your AI-powered workflow.
Create tasks, drag them between columns, and see at a glance what every agent is doing. Each task is a unit of work — a feature, a bugfix, a refactor. You stay in control of the big picture.
Every task on the board gets its own isolated git worktree. Inside that worktree you can run as many agents as you want — in parallel, same codebase, without stepping on each other’s toes. Claude, Cursor, Codex, Gemini — all at once on the same task if needed.
Each task opens a real terminal powered by Ghostty’s rendering engine with tmux preconfigured. Split panes to run several agents side by side in the same worktree — Claude, Gemini, Codex, whatever you need. Scroll history, copy-paste, all native features intact.
Every feature is designed around one goal: managing many AI agents across many tasks without losing your mind.
Hover over any task card on the Kanban board to see a live preview of its terminal — without switching views. Instantly check what each agent is doing. The preview updates in real-time, every second.
Full repo isolation. No merge conflicts between parallel agents. Each task works on its own branch and copy.
Each task launches with tmux and your preferred command. Split panes, persistent sessions, full terminal power.
Agents send a bell when they need your attention. Red badges appear on task cards so you never miss a prompt.
Move tasks between status columns with drag-and-drop. Visual feedback, smooth animations, intuitive flow.
Organize tasks with colored labels (Bug, Feature, Refactor…) and find anything instantly with full-text search. Create tasks from remote branches — just paste a branch name, hit Fetch, and the agent starts working on it.
Don’t know exactly how something should look? Run the same task 3, 5, or 10 times — with different agents, models, or prompts. Each attempt gets its own isolated worktree. Browse the results, keep the winner, delete the rest.
Check out any remote branch and toggle “PR review” to pre-fill a structured code-review prompt. The agent reviews — not modifies — the code.
Configure a setup script per project. Install deps, seed data, start services — runs automatically for each new task.
Launch your app straight from the task’s worktree with a single click. Restart or stop instantly — no terminal juggling. Configure the dev command per project and every task gets its own isolated server, ready in seconds.
Hit ⌘/ to open a single reference for
every keybinding — app navigation,
quick-switch and zoom on the App tab, and the
complete tmux keymap (panes, layouts, windows,
sessions, copy mode) on the Terminal tab.
Stop memorizing — just look it up.
Define your own pipeline: AI Review, PR Review, On Hold — whatever stages your team needs. Move tasks through your custom flow with a click.
Full theme support. Switch between dark and light modes to match your environment and preference.
Clone node_modules, .venv, and other heavy directories into worktrees instantly with near-zero disk usage. Auto-detect finds well-known paths across 12+ ecosystems.
Set a port count per project and every task gets its own free ports, injected as $DEV3_PORT0, $DEV3_PORT1… into your scripts. No more port clashes between parallel dev servers.
Press ⌘⇧N to open another dev-3.0 window and put a different project on each screen. Close any window without quitting the app.
Give any column an agent that auto-spawns the moment a task lands in it — plus a per-column instruction telling the AI exactly when to move work there. Your board runs itself.
Completed tasks pass through a configurable AI reviewer before they reach you. Pick the review agent, model, and prompt — it catches bugs and fixes the small stuff first.
Working in a giant monorepo? Sparse-checkout just the directories each task needs — worktrees create faster and use a fraction of the disk.
AI writes the code now — your job is to command the fleet. The bottleneck moved from your editor to your focus. Two things we optimize for, above all.
dev-3.0 optimizes a single developer: you. The unit is always the individual, never the org. Use it on a team if you like — but it’s not for managing other people; it’s for each person to command their own fleet at their own top speed.
A cockpit you stare at all day should be fast, gorgeous, and keyboard-first — and bend to your way of working, not force one on you. Great tooling doesn’t just make you productive; it makes the work fun again. We sweat the polish.
And what we refuse: dev-3.0 is not an IDE — and won’t become one. Code and git are the agent’s job. Need an editor? One click to your real VS Code or Cursor — the goal is to need it less. Integrations? Your agent already speaks MCP to Linear, Jira and the rest. dev-3.0 is the cockpit; your agent is the adapter.
Free, open source, and climbing. Join the developers who already command their agent fleets with dev-3.0 — and star the repo to keep the curve going.
Everyone else is building an editor for agents. We’re building mission control for you. Pick by your goal, not a feature checklist.
Hands on the editor and git, reviewing every line yourself? An agent IDE will feel like home.
Need SSO, seats, audit logs and a managed platform? A team orchestrator fits better.
AI writes the code; your job is running the fleet at full speed without drowning. That’s dev-3.0.
dev-3.0 is a native app for macOS and Linux. The fastest way in: paste one line into the coding agent you already use — it installs everything for you.
Paste this into Claude Code, Codex, Gemini CLI — whatever you already run. The agent reads the guide, detects your OS, and does the whole install itself.
Install dev-3.0 by following the guide at https://dev3.h0x91b.com/ai-install.txt
Machine-readable docs: ai-install.txt · llms.txt
brew tap h0x91b/dev3 && brew trust h0x91b/dev3 && brew install --cask dev3
Apple Silicon & Intel. Prefer a classic install? Grab the DMG: Apple Silicon · Intel
brew tap h0x91b/dev3 && brew trust h0x91b/dev3 && brew install h0x91b/dev3/dev3
dev3 remote
Prints a QR + URL — drive the full UI from any browser, even your phone.
Desktop GUI via dev3 gui.
No Homebrew? CLI tarball.
Your agent installs the app and its dependencies
Point it at any git repo you already have
Describe the work — your first agent is running
macOS (Apple Silicon & Intel) and Linux — Windows coming soon
Free & open source (Apache 2.0) • Upgrade anytime: brew trust h0x91b/dev3 && brew upgrade --cask dev3
dev-3.0 doesn’t need your laptop open. Run it headless on any Linux box and command your agents from wherever you are.
dev3 remoteStart it on a cloud VM or home server — it prints a QR code and URL. Scan with your phone or open in any browser: the full UI, boards, live terminals, everything. A built-in Cloudflare tunnel means no SSH gymnastics, and agents keep working while you’re away. Check on the fleet from the couch, approve a review from the train.
dev-3.0 doesn’t impose a flow. Size your tasks and arrange your space around how you actually work. A few of the ways people run it:
One task, one worktree; spin up many in parallel and let agents run to done. Size each task so you can verify it in ~90 seconds — if you can’t, it’s too big. AI review, one-click dev server and live preview make checking instant.
Barely writing code? Fan many tasks — each with several variants — at a question: measure, analyze, search. Then a roll-up task merges the findings, surfaces conflicts, and spawns the next round. A whole investigation, parallelized.
Branch off your own branch instead of main, then knock out small, isolated sub-tasks — each trivial to verify — that merge back up. The big branch assembles itself into your feature, with near-zero cognitive load per step.
…and more — multiple windows, multiple boards, many projects, many agents per task. Arrange it however fits you.
dev-3.0 doesn’t integrate with agents — it hosts them. Any tool that runs in a terminal works out of the box. When agents ship new features — like Claude’s multi-agent teams — you get them instantly, no update needed.
Built on a native webview runtime, GPU-accelerated terminal, and the fastest JS runtime available. Instant startup, minimal memory footprint, and a silky-smooth experience.
Native macOS webview runtime. Lightweight, fast startup, tiny memory footprint.
All-in-one JS runtime. Fast installs, native APIs, built-in bundling. Main process runs in Bun.
GPU-accelerated terminal rendering engine. Buttery smooth, true-color, fast as native.
Battle-tested terminal multiplexer. Persistent sessions, split panes, keyboard-driven workflow.
Dozens of small touches that add up. Here are a few favorites — the full list is one click away.
Hit ⌘⇧P, type a few letters (theme, dev server, pull main…), Enter. Every action, no menus.
Launch one TODO as several variants — each gets its own agent, worktree and terminal. Compare, keep the best.
Just say “create a task in dev3 about…” in your prompt and it lands on your board.
Mention “dev3” and the agent can list tasks, check statuses, and dig up past work.
Agents record key decisions and findings as task notes — context survives long after the worktree is gone.
Drop a task in the AI Review column and an agent reviews your branch, fixing the medium/high issues.
⌘V or drag an image into a task or note — saved automatically and shown as a thumbnail.
Select text in any terminal pane and it’s on your clipboard instantly — works over SSH too.
In Show Diff, click + next to any line to leave an inline note exactly where it matters.
Reopen a finished task and the previous AI conversation picks up instead of starting cold.
Hold Option, tap Tab to flip through active tasks with live previews; add Shift to span every project.
Save scripts and paths to .dev3/config.json — teammates clone the repo and get the whole setup.
Every shortcut lives in one place. Inside the app, press ⌘/ (Ctrl+/ on Linux) — or open Help → Keyboard Shortcuts — to see them all (App + Terminal tabs).
| Action | macOS | Linux |
|---|---|---|
| Go to project (quick switch) | ⌘K | Ctrl+K |
| Command palette | ⇧⌘P | Ctrl+Shift+P |
| Keyboard shortcuts panel | ⌘/ | Ctrl+/ |
| Back / Forward | ⌘[ / ⌘] | Ctrl+[ / Ctrl+] |
| Switch to project 1–9 (keep view) | ⌘1–9 | Ctrl+1–9 |
| Switch to project 1–9 (flip view) | ⇧⌘1–9 | Ctrl+Shift+1–9 |
| Cycle active tasks (this / all) | ⌥Tab / ⌥⇧Tab | Ctrl+Tab / Ctrl+Shift+Tab |
| New task / Add project / New window | ⌘N / ⌘P / ⇧⌘N | Ctrl+N / Ctrl+P / Ctrl+Shift+N |
| Settings | ⌘, | Ctrl+, |
| Zoom in / out / reset | ⌘= / ⌘- / ⌘0 | Ctrl+= / Ctrl+- / Ctrl+0 |
| Toggle project terminal / open Quick Shell | ⌘` / ⇧⌘` | Ctrl+` / Ctrl+Shift+` |
| Hard refresh | ⌘R | Ctrl+R |
| Close dialog / step back | Esc | Esc |
Terminal multiplexing uses tmux's ⌃B prefix bindings — shown on the Terminal tab of the same panel.
dev-3.0 is free and open source. Star the repo, build from source, start shipping.