dev-3.0
Run coding agents in isolated worktrees and terminals, then track and review their work from one board.
- Source repo
- h0x91b/dev-3.0
- Stars
- ★ 324
- Last updated
- 1d ago
- License
- Apache-2.0
- Primary language
- TypeScript
- FA score
- 76/100 · Good
At a glance
- How it runs
- Works with
- Universal · cross-platformCodex · Claude Code
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- A solo developer wants multiple coding agents to work on separate tasks in one repository without sharing a worktree.
- Not a fit if
- Developers who need an embedded editor or manual Git staging and commits
- Teams seeking centralized member management, SSO, and seat administration
- Source review
- 76/100 · Good
What does this agent do, and when should you use it?
dev-3.0 is a desktop task board for individual developers, with each card representing a live AI coding task. Starting a task creates a separate Git worktree and tmux session, and can run a project setup script and reserve ports for a development server. The board provides terminal previews, side-by-side agents, task variants, and a built-in diff panel for inline review comments. It supports Claude Code, Codex, Gemini CLI, Cursor Agent, opencode, and configurable command-line tools. Install it as a macOS or Windows desktop app, or run it headlessly on Linux and open its interface in a browser.
Add a Git repository as a project, create a task, and choose one or more agents; dev-3.0 creates the task's Git worktree and tmux session, then launches the selected agent. The board shows agent status, questions, priority, and task progress, with terminal previews; agents can run in split panes for one task, or as variants on separate branches. Review completed work in the diff viewer, comment on changed lines, and copy the review back to the agent's terminal. Agents can also use the dev3 CLI to update task overviews, leave notes, request attention, show images or HTML reports, start related tasks, and inspect another task's terminal read-only. The app can display PR numbers and CI status.
- A solo developer wants multiple coding agents to work on separate tasks in one repository without sharing a worktree.
- A developer wants to compare alternative implementations by running task variants on separate branches and reviewing their diffs.
- A developer needs to keep an eye on agent progress and questions from a board instead of switching into each terminal.
- A developer wants to review generated code file by file, leave inline feedback, and send that feedback back to the agent.
- A developer runs agents on a headless Linux machine and wants to access the board and terminals from a phone or browser.
How do you install or deploy this agent?
On macOS, install with Homebrew:
brew tap h0x91b/dev3
brew trust h0x91b/dev3
brew install --cask dev3For a headless Linux setup, install the CLI and start remote access:
brew tap h0x91b/dev3 && brew trust h0x91b/dev3 && brew install h0x91b/dev3/dev3
dev3 remoteOn Windows, download stable-win-x64-dev-3.0.zip, extract it, and run bin\\launcher.exe. The install guide also documents a direct DMG, CLI tarball, systemd service, and building from source. The supplied README does not state the product price or whether agent accounts incur charges.
How do you use this agent?
Launch the app and add a Git repository as a project. Create a task, describe the work, select an agent, and press Run. The agent runs in a task-specific worktree and tmux session; follow it from the board or terminal preview, then review its changes in the built-in diff panel. On a headless Linux machine, start the browser-accessible UI with:
dev3 remoteAgents can interact with the board through the CLI, for example to request attention or inspect a peer task's terminal read-only:
dev3 attention "Need a decision on the schema"
dev3 peek --task seq:1383What are this agent's strengths and limitations?
- A separate Git worktree and tmux session per task isolates concurrent agents' working files.
- The built-in diff viewer, inline comments, and review handoff provide a concrete workflow for checking agent output.
- Supports agents in split panes, branch variants for a task, and customizable board columns.
- Offers headless Linux operation with browser-based remote access, including access from a phone.
- It explicitly has no embedded code editor and does not center manual staging or hand-written commits; users can open VS Code or Cursor when needed.
- It relies on Git, tmux, and shell-based coding agents; status reporting, session resume, and other features vary by agent.
- Windows support is described as brand new and potentially rough, and its current installer is unsigned.
- Telemetry is enabled in official release builds; users concerned about it need to understand the collected data and opt-out controls.
How does this agent compare with similar options?
The README distinguishes dev-3.0 from agent IDEs and team platforms: choose an agent IDE to edit code and manage Git inside an editor, or a team orchestrator for features such as SSO, seats, and audit. dev-3.0 is aimed at developers coordinating their own agent fleet.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| dev-3.0 This agent | 76 · Good | Desktop app | ★ 324 | 1d ago | TypeScript | Codex · Claude Code |
| Agent Deck | 76 · Good | CLIFree + model costs | ★ 1.2k | today | Go | Codex · Claude Code |
| Agent Manager | 64 · Some gaps | CLIFree | ★ 633 | today | Go | Codex · Claude Code |
| Diri | 59 · Major gaps | Desktop appFree + model costs | ★ 483 | today | Rust | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
The README describes GA4 and PostHog, opt-out controls, collected fields, and explicit exclusions such as project names, prompts, diffs, and terminal output, so data-flow transparency is strong. The docs also specify cleanup scope, protection for unmerged work, and how to revoke Full Disk Access. Tasks can access repositories and launch configurable shell agents, while remote access can use tunnels or LAN; permission and external-effect boundaries are not fully established, so those scores are not full. Credential-token risk is called out, but the supplied material does not demonstrate comprehensive protection. Dependency security has version pins and install guidance but no dependency-audit evidence. Authorship and licensing are traceable in the repository; publisher identity is unverified, so maintenance responsibility receives only partial credit.
Install and troubleshooting docs identify tmux version issues, platform differences, Windows update uncertainty, and diagnostic commands, providing useful availability and failure guidance. Some support and fix descriptions are stated as known facts, but this is a static review only. The product marketing name and package metadata name also differ, so consistency is only partial.
The material clearly targets solo developers and distinguishes the product from IDEs and team orchestrators. It documents desktop, headless Linux, WSL, Windows, and agent-support differences. Custom columns, multiple agents, and remote topologies offer useful adaptation. A bound agent can start when a card is moved to its column, but conditions and configuration limits are not fully explained; platform support is detailed but varies in maturity, so environment fit is not full.
README, installation, remote access, troubleshooting, support matrix, and exit-code docs have clear roles. Install paths, prerequisites, updates, removal, and platform-specific limitations are concrete. Examples and FAQ-like troubleshooting are extensive, and the docs disclose Windows, telemetry, and permission limitations. The full Apache-2.0 text and copyright are present. Version updates rely on a generated changelog and release page, but the supplied material does not include a complete history; the repository identifies an author, while independent maintenance commitments and response responsibility remain unclear.
The board, isolated worktrees, terminal previews, diff review, and agent notifications directly support parallel coding-agent management. Outputs are reviewable and feedback can be sent back to agents. The docs describe concrete value over manual terminal switching, but provide no quantified resource, scale, or cost-benefit evidence, so cost-benefit receives partial credit.
Important README claims about telemetry, cleanup, and platform limitations point to specific settings, commands, or dedicated docs. Install and troubleshooting materials complement each other, and package.json and LICENSE support checks of dependency, license, and project metadata. The supplied material includes code and CI references, but not implementation source for item-by-item corroboration, so cross-source support is partial. The docs generally distinguish current behavior, observed findings, and unresolved items; this review also makes no execution claim, so fact-inference separation is strong.
- Publisher identity is not verified by the enterprise registry; independently verify release provenance and protect plaintext login credentials in agent account directories.
- Remote access may expose the UI through a tunnel or LAN. The install guide says the default bind address is 0.0.0.0; explicitly restrict the host or disable tunneling when appropriate.
- Windows support is early and builds are unsigned. Linux tarball users must manage tmux versions themselves; the docs specifically advise avoiding 3.7.x.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
Release notes excerpt
What's new in v1.62.2 Fixes - A board whose initial task fetch was lost (for example across a remote WebSocket reconnect) no longer sits on "loadin... - A second dev3 instance on the same machine (for example a dev build started from a task) no longer deletes the worktr... - Settings are never saved from the screen's pre-load placeholder any more: when the first settings read at launch was ... - Completed and cancelled tasks no longer match pr:open or show an open, mergeable PR badge from a status snapshot fr... Download macOS (Apple Silicon) …View on GitHub ↗ -
Release notes excerpt
What's new in v1.62.0 Features - Each task can now have one Live Artifact: a page that updates in place instead of adding versions, with its own small... - Every task now records who created it — you, another task (in any project), an automation or a conversation import — ... - dev3 now runs dev3 doctor --worktrees --prune-older-than 14d by itself once a day, never in the first hour after la... - Terminal tabs above a task's terminal can now be renamed in place: double-click a tab, click the pencil that appears ... …View on GitHub ↗ -
Release notes excerpt
What's new in v1.61.0 Features - Clicking an audio or video file link in a terminal now plays it inside the file preview with native controls (play/pa... - Agents can now park and revive tasks: dev3 task hibernate does what the session bar's Hibernate button does, and `d... - The Add Project dialog has an optional Auto-configure switch that reads a new project's lockfiles and package.json an... - dev3 dev-server start|restart [server] -- <args...> appends the arguments to the server's script, each shell-quoted... …View on GitHub ↗
FAQ
Is this a code editor?
Which coding agents does it support?
Can I use it on a headless Linux machine?
dev3 remote for remote access.