Run AI coding agents side by side, each in its own git worktree

What is daruda

An Agentic Development Environment

daruda is a desktop app for running several AI coding agents at once without juggling terminal tabs and branches by hand. Each task gets its own git worktree, so agents can work on separate changes in one repository at the same time. daruda is not a coding agent itself: you bring your own subscriptions and API keys, agents run on your machine, and daruda does not proxy model calls.

Features

Supervise many agents without losing track

Review every tool call

Each step an agent takes is a card: commands, file reads and edits, web fetches, and MCP calls, each with its status and duration. Expand one to see its input and output.

Answer agents in place

When an agent asks a question or requests permission, the prompt appears in the transcript with its options, and the input dock switches to answering it.

Run flows and inspect every node

Flows chain agent and command steps. The run view shows the graph, each node's status and attempts, and the saved conversation behind it.

Delegate to the orchestrator

The orchestrator is a resident agent you reach with /daruda. Ask it to create worktrees, start tasks and flows, or message other agents. Before it creates a worktree, chat, or task, or starts a task, it waits for your approval as a card in its chat.

Reach agents from Telegram, Slack, or Discord

Remote Control connects a bot so you can message the orchestrator and your agents from your phone, receive their results, and answer their approvals. Turn on Only while away to forward messages only when you are away from daruda.

Terminals in the same window

Terminal panes use the Ghostty terminal engine. Split them beside agent chats for shells, CLI agents, and one-off TTY work, with IME input, scrollback search, and prompt marks.

• daruda as an MCP server: register daruda --mcp with any MCP client so agents can create worktrees and chats, run tasks and flows, and wait for results while daruda is running.
• Session hosts: run a worktree's agents on an SSH host or in a Docker container.
• Multiple accounts: sign in to several Claude and Codex accounts and pick one per pane, with credentials kept apart.
• Usage limits: rate-limit windows in the status bar, and an optional stop before a reply would bill extra usage.
• Keep awake: keep the computer awake while agents work or wait for an approval.
• Notifications: get notified when an agent finishes, a long command ends, or a terminal agent stops to ask you something.
• Chat details: queued prompts, file and image attachments, subagent activity, and the agent's plan.
• Workspace views: Markdown, Mermaid, and image previews; git changes; tasks; skills; MCP tools; per-worktree ports.
• Tasks: write a task once, then start it in a new or existing worktree.
• Make it yours: dark and light themes, separate fonts, remappable shortcuts, status icon plugins, and an English or Korean UI.

Supported agents

Bring the agents you already use

daruda ships ACP presets for 38 agents. Install and enable them from Settings → Agents & Chat. Any other CLI agent runs in a terminal pane.

• Claude Agent
• Codex
• Gemini CLI
• GitHub Copilot
• Cursor
• OpenCode
• Amp
• Factory Droid
• Devin
• goose
• Kimi CLI
• Qwen Code
• Mistral Vibe
• Cline
• Junie
• Kilo
• Auggie CLI
• Grok Build
• and 20 more

Install

Download daruda

Every build is on GitHub Releases. You also need Git and at least one coding agent you are signed in to.

xattr -cr /Applications/daruda.app

Start with one task

1. Click Add Project and choose a repository on your computer.
2. Hover over the project header, click +, and name a branch to Create Worktree for the task.
3. Open New Agent Chat, choose an agent, and describe a small change. Follow its tool calls, answer its questions, and review the diff before you commit.