• Tiled panel layout with drag-to-reorder
• Focus mode — single-task layout with a clean two-column view on wide screens (Ctrl+Shift+F)
• Built-in diff viewer with inline review comments and per-commit navigation
• Guided change tour — click Generate tour of changes at the bottom right of a task's Changed Files panel to generate in the background while you work. Automatic scope includes the whole feature branch against its configured or detected base, including uncommitted work; on main, master, and develop, it includes only uncommitted changes. Choose Current diff to explicitly use the commit selector (All, Uncommitted, or a single commit). Uncaptured context stays omitted when reading a captured tour. Hover or focus the button for an explanation and the configured provider/model. The button shows a spinner and progress, then a green checkmark and Start tour when ready. Click Start tour to open the diff captured for the tour; it never opens automatically. Completed tours remain available when you close and reopen the tour; clicking the loading button cancels generation. Uses your configured code Q&A provider, aiming for 4–6 concise stops in one request with a dedicated 500,000-character prompt budget; inline Q&A keeps its 50,000-character limit. This is an app guard, not an exact token limit—provider context limits still apply. Larger diffs use compact excerpts and a five-minute timeout per part. Claude tour prompts are piped through stdin to avoid command-line size limits. Verbose logs include request sizes, time to first output, and total request duration, without code or response contents. Tours describe the supplied diff and do not verify test execution. Each stop is a card with a role label and a tone that flags risky or uncertain changes; Ask and Go deeper answer follow-ups under a stop with the diff as context. Jump between stops from the progress strip, or with the arrow keys and Home/End while the tour has focus. The button shows Resume tour when you left midway.
• All tours — click a progress segment to jump to that card (a dot marks cards with answered follow-ups), copy the tour with its answers as Markdown, or Rework it with your own instructions ("focus on error handling", "for a newcomer"), which regenerates it from the same context. Code references open at their line when your editor command is VS Code, Cursor, VSCodium, Windsurf, Zed or Sublime Text; other editors open the file.
• Understanding tours — Take Tour beside Review Plan (and in the plan viewer's header) explains the plan; Understand on a changed-file row explains that file. Both open a short card tour: a gist, then 3–7 cards with Ask and Go deeper, which generate a small branch you return from. They use the same code Q&A provider setting as the change tour and inline Q&A, and run without tools: a plan tour sends the plan markdown, a file tour sends the file plus its direct imports (size-capped; anything omitted is listed on the gist card). Tours live in memory for as long as the task panel is open and are discarded when you switch tasks. A reopened file tour whose file has changed shows a Regenerate notice instead of silently going stale. Free-form subsystem questions ("how does X work?") are not supported yet.
• Task brief — after each agent turn, a cheap model (the task-naming model) rewrites a short brief from the turn's prompts, the agent's last message and the branch diff: what changed, changes beyond the goal, what is still open, and decisions the agent made on its own. A card above the agent's output shows the headline; click it for the full brief, where you can keep a decision or send the agent a correction. Off by default; turn it on under Settings → Write task briefs
• Confidence ratings — Rate confidence on the brief card (shown on every task, with or without briefs) asks the agent to rate its work and publish it with the confidence_publish MCP tool. Scores are named by whoever rates (one plain score or your own dimensions), low ones are highlighted, and a rating is marked out of date once the code changes after it. Point the prompt at your own rating skill in Settings
• Notes panel per task — jot ideas, then send the notes straight to the agent as a prompt
• Canvas per task — a Markdown file from the worktree rendered live as the agent writes it; edit it in place, or select a passage and send it to the agent with your question
• Browser preview in the canvas — run a local app in the task shell, open + → Browser, and pick elements to reference in your prompt (usage and limits)
• PR CI status watcher — desktop notification when GitHub checks settle
• Super Productivity integration — the task you focus is tracked in Super Productivity (a matching task is created there on first focus, in the Super Productivity project you link in project settings), titles stay in sync, and merging or closing a task marks it done there. A task Super Productivity is already tracking that isn't a Parallel Code task is never replaced — a banner offers to switch instead. Connect in Settings with the token from Super Productivity's Local REST API (Settings → Misc; the API must stay turned on). With Super Productivity's bundled Parallel Code plugin enabled, Start in Parallel Code on a task opens a pre-filled New Task form here; that link needs the macOS app or the .deb (an AppImage only handles it if it was integrated into your desktop).
• Shell terminals per task, scoped to the worktree
• Direct mode for working on the main branch without isolation, plus support for folders without a git repo
• Existing worktree import — bring already-created worktrees into Parallel Code
• Sandboxing with project-specific Dockerfiles — drop a .parallel-code/Dockerfile into the project and tasks run inside it
• Coverage radar — per-file test-coverage badges in the Changed Files panel
• Configurable keyboard shortcuts with per-agent presets
• 10 themes — Islands Dark, Minimal, Graphite, Midnight, Classic, Indigo, Ember, Glacier, Zenburnesque, Workbench
• State persists across restarts
• macOS and Linux

Antigravity (agy) signs in interactively and caches credentials in your OS keyring (Keychain/libsecret). The keyring cannot be reached from inside a Linux container — it needs a secret-service daemon the agent container doesn't run — and agy has no API-key fallback, so Docker-isolated Antigravity tasks cannot authenticate. Run Antigravity as a native (non-Docker) task.

The bundled image still ships the agy binary, and ~/.gemini/antigravity-cli (settings/plugins) is shared when "Share agent auth" is enabled, so Docker support is forward-compatible if a file-based or API-key auth path appears later.

git clone https://github.com/johannesjo/parallel-code.git
cd parallel-code
npm install
npm run dev

Requires Node.js v18+.

Phone access uses port 8777 in development (npm run dev) and 7777 in the installed app, so both can run together. Run npm run build:remote after changing the mobile UI, then use Connect Phone in the development app to get its connection link or QR code.

Ctrl = Cmd on macOS.