Parallel Code: Run Claude Code, Codex, and Gemini in Isolated Git Worktrees
Run Claude Code, Codex, and Gemini side by side, each in its own git worktree.
At a glance
- What is it?
- Parallel Code is an open-source desktop app that dispatches multiple AI coding agents side by side, each in its own git worktree. It targets developers who want parallel agent runs without manual branch and merge management.
- Who is it for?
- Adopt Parallel Code if you regularly run multiple AI coding agents and want automatic git isolation, a GUI for diff review, and keyboard-first control, especially on macOS or Linux. Do not adopt it if you need Windows support, require a web-based interface, or rely on Antigravity CLI in Docker, which cannot authenticate due to keyring limitations.
- Can I use it commercially?
- Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository last received commits 1 day ago.
- What is it written in?
- Mainly TypeScript, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 14, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
The Problem: Parallel Agents Without Manual Git Chaos
Running AI coding agents sequentially blocks your workflow. Running them in parallel across multiple terminal windows or tmux panes works, but you manage worktrees, branches, and merges by hand. Parallel Code solves this by giving each agent its own git worktree automatically. The README's comparison table is blunt: multiple terminal windows have no GUI and no automatic git isolation, while VS Code extensions are tied to VS Code and lack true parallel worktree isolation. This tool is for developers who already use Claude Code, Codex, Gemini, or Copilot CLI and want to dispatch several agents at once without tripping over branch conflicts. The pitch is concrete: ten agents, ten branches, one afternoon. If you are a solo developer or a small team that juggles many small feature tasks, this is the target audience.
How It Works: Worktrees, Symlinks, and Spawned Agents
The core mechanism is straightforward. When you create a task, Parallel Code creates a new git branch from your main branch, sets up a git worktree so the agent works in a separate directory, symlinks node_modules and other gitignored directories into that worktree, and then spawns the agent CLI in the worktree. The symlinking step is critical: without it, each worktree would need a fresh npm install, which would slow down parallel runs. The README does not specify whether symlinks are created for all gitignored paths or only node_modules, so there is some ambiguity. Once the agent finishes, you merge the branch back to main from the sidebar. This design means every change is isolated and reviewable, and you can run five agents on five features simultaneously with zero conflicts. The data flow is simple: task creation triggers branch creation, worktree setup, symlinking, and agent spawn. The GUI then tracks each session, and you review diffs before merging.
Getting Started: Download, Install an Agent CLI, and Point at a Repo
Installation is minimal. You download the latest release for macOS (.dmg, universal) or Linux (.AppImage or .deb) from the releases page. You must have at least one AI coding CLI installed: Claude Code, Codex CLI, Gemini CLI, Antigravity CLI, or Copilot CLI. Then you open Parallel Code, point it at a git repository, and start dispatching tasks. For building from source, the README gives three commands: git clone, npm install, and npm run dev. It requires Node.js v18 or higher. There is no mention of Windows support, and the README explicitly says macOS and Linux. The keyboard-first control is a selling point: every action has a shortcut, and Ctrl equals Cmd on macOS. The README lists shortcuts like Ctrl+N for new task, Ctrl+Enter to send prompt, Ctrl+Shift+M to merge to main, and Ctrl+Shift+P to push to remote. The setup is genuinely low-friction, but you must already have the agent CLIs installed, which is an external dependency.
The Antigravity Docker Limitation: A Real Failure Mode
The README includes a detailed warning about Antigravity CLI. Antigravity signs in interactively and caches credentials in your OS keyring (Keychain on macOS, libsecret on Linux). That keyring cannot be reached from inside a Linux container because the agent container does not run a secret-service daemon. Antigravity has no API-key fallback, so Docker-isolated Antigravity tasks cannot authenticate. The README explicitly states: run Antigravity as a native (non-Docker) task. This is a genuine limitation that could trip up users who assume all agents work identically in sandboxed environments. The bundled image still ships the agy binary, and ~/.gemini/antigravity-cli settings and plugins are shared when Share agent auth is enabled, so Docker support is forward-compatible if a file-based or API-key auth path appears later. But right now, if you want to use Antigravity in a Docker-isolated task, it will fail. This is the kind of edge case that matters when you are evaluating whether the tool fits your stack.
Features That Go Beyond Basic Worktree Management
Parallel Code is not just a worktree wrapper. The README lists a tiled panel layout with drag-to-reorder, a focus mode for single-task layouts, a built-in diff viewer with inline review comments and per-commit navigation, a steps tracking panel that writes to .claude/steps.json, per-task notes that can be sent to the agent as prompts, a PR CI status watcher that notifies when GitHub checks settle, shell terminals per task scoped to the worktree, direct mode for working on the main branch without isolation, support for folders without a git repo, existing worktree import, sandboxing with project-specific Dockerfiles, coverage radar with per-file test-coverage badges, configurable keyboard shortcuts with per-agent presets, and ten themes. The coverage radar is interesting: it shows per-file test-coverage badges in the Changed Files panel, which suggests the tool integrates with test coverage tooling, though the README does not specify which coverage tool. The steps tracking panel writes to .claude/steps.json, which implies Claude Code integration is deeper than others. The phone monitoring feature is notable: scan a QR code and watch agents work over Wi-Fi or Tailscale. This is a practical feature for long-running agent tasks.
Maintenance and Upgrade Cost: Release Cadence and License
The repository shows a recent release cadence: v1.14.5 on 2026-08-21, v1.14.3 on 2026-08-18, and v1.14.2 on 2026-08-13. That is multiple releases per week, which indicates active maintenance. The project is MIT licensed, so you can use, modify, and distribute it freely, with no extra platform fee. The README emphasizes that it is free and open source, with no extra subscription required beyond the agent CLIs themselves. However, the maintenance cost for you as a user includes keeping up with frequent releases, which may introduce breaking changes or new features. The README does not mention an upgrade path or migration notes, so you should check release notes before upgrading. The state persists across restarts, which reduces the cost of switching between sessions. The license implications are straightforward: MIT allows commercial use, but you are responsible for the agents you run, and the tool does not include any agent subscriptions. If you rely on a specific agent version, you need to verify compatibility with each Parallel Code release.
Alternatives: Terminal Windows vs. VS Code Extensions vs. Sequential Runs
The README itself names the alternatives: multiple terminal windows or tmux, VS Code extensions like Kilo Code and Roo Code, and running agents sequentially. The difference is in the approach. Terminal windows give you full control but require manual worktree and branch management, which is error-prone. VS Code extensions are integrated into the editor but are tied to VS Code and do not provide true parallel worktree isolation between agents. Sequential runs are simple but block your workflow while each agent finishes. Parallel Code's approach is to automate the isolation and provide a GUI for review, which is a middle ground between raw terminal control and editor-specific extensions. If you are already comfortable with tmux and git worktree commands, you may not need Parallel Code. But if you want a visual diff review and a unified session view, it offers something the terminal approach does not. The comparison is honest, but it does not mention any terminal multiplexer enhancements like tmux resurrect or tmuxinator, so you should evaluate your own workflow.
Editorial conclusion
Adopt Parallel Code if you regularly run multiple AI coding agents and want automatic git isolation, a GUI for diff review, and keyboard-first control, especially on macOS or Linux. Do not adopt it if you need Windows support, require a web-based interface, or rely on Antigravity CLI in Docker, which cannot authenticate due to keyring limitations. Before adopting, verify that your preferred agent CLI is installed and that your project's gitignored directories (like node_modules) symlink correctly into worktrees, as the README does not detail failure modes for symlinking. Also confirm that your workflow benefits from the worktree model, since direct mode exists but bypasses isolation.
Community notes