What is Orca? Run multiple coding agents in parallel, each agent one git worktree

October 20, 2026

Updated:

Guide
What is Orca? Run multiple coding agents in parallel, each agent one git worktree

What is Orca? Orca is an open-source application that helps you run multiple coding agents at the same time, Claude Code, Codex, OpenCode, Pi, Oh My Pi and dozens of other agents, with each agent working in its own git worktree, and you monitor everything on one screen. If you're used to handing work to an agent and then sitting around waiting, Orca lets you hand five tasks to five agents and only come back when an agent needs you.

Orca on a computer running multiple agents in parallel, the Orca app on a phone in the corner
Orca on desktop: list of worktrees on the left, running agents in the middle, accompanying mobile app in the top right corner. Image: official project README.

What is Orca?

Orca is an open-source application developed by Stably, released under the MIT license, used to orchestrate multiple coding agents running in parallel. The development team calls this an agent development environment (ADE): instead of the interface revolving around a code writer's cursor, the interface revolves around a team of agents. Orca doesn't write code itself. The application opens terminals, creates worktrees, starts the agents you choose, notifies you when an agent is done and helps you review changes. Claude Code or Codex underneath still use your own account, so Orca is free while AI costs belong to the agent provider.

According to GitHub statistics, read on 19/10/2026: the project launched in March 2026, has around 96,000 stars, the latest version is 1.4.229 released on 17/10/2026. Orca runs on macOS, Windows, Linux and has iOS and Android apps.

In practice, the README shows the development team updates daily, so the real list of features is in the changelog on GitHub. For example: you assign 5 independent tasks and each agent runs for 20 minutes. Running in parallel, the whole batch finishes in about 20 minutes instead of 100 minutes when done sequentially, not counting the time you review the results.

Why run agents in separate worktrees?

Separate worktrees are necessary because two agents modifying the same directory will overwrite each other: agent A just edited a file and agent B reads the old version and writes it back. A git worktree is a separate working directory of the same repo, on its own branch, so each agent works and tests separately, while the main branch stays untouched. You can create one yourself using git worktree add, but with five agents at the same time, the tedious part is remembering which agent is where, whether it's done, what it changed. Orca handles that management part.

The official README documentation for Orca (October 2026 version) lists two typical usage patterns. Pattern 1 is dividing one request among five agents, each agent in one worktree, comparing results and merging the best version. Pattern 2 is assigning five different tasks to five agents and handling them one by one as each agent finishes.

Example: you need to add a login page to a Next.js project. First, assign the task to Anthropic's Claude Code and OpenAI's Codex in two worktrees. Then open the diff from each side, about 10 minutes each, total 20 minutes, then choose the more concise version.

Split-pane terminal in Orca
Split-pane terminal in Orca (keyboard shortcut ⌘⇧D to split). Image: official README.

What are Orca's main features?

The core point of Orca is bringing all agents into one management screen. The list below follows the official README, read on 19/10/2026. First, nine main feature groups:

  1. Parallel worktrees: each agent in an isolated worktree, tracked in the sidebar.
  2. Split-pane terminal: like Ghostty, drawn with WebGL, unlimited splitting.
  3. Design Mode: click a UI element in the Chromium window, Orca sends HTML, CSS and screenshot into the prompt.
  4. Diff annotations: add notes to each line the AI changed and send back to the agent.
  5. GitHub and Linear: view PRs, issues and open worktrees from a task.
  6. Worktrees over SSH: run agents on a powerful remote machine, auto-reconnect.
  7. Mobile app: receive notifications when an agent finishes.
  8. Drag files into agent: edit like VS Code, drag files or images into the prompt.
  9. Usage tracking: see Claude usage from Anthropic and Codex from OpenAI, switch accounts without re-logging in.

For example, when editing UI, you use Design Mode to select a button and send that button's code to the agent, avoiding describing it in words.

Design Mode: select UI element to send into agent prompt
Design Mode running on the Chromium browser embedded in Orca. Image: official README.

Annotate each line in the AI-generated diff
Diff annotations: add notes to code lines and send back to agent. Image: official README.

Which agents does Orca support?

Orca supports any agent that runs in the terminal, meaning you're not locked into a single provider. The README I read on October 19, 2026 put it briefly like that, and the available list with icons includes over 30 agents: Claude Code, Codex, Cursor CLI, GitHub Copilot CLI, Grok, OpenCode, Antigravity, Pi, Oh My Pi, Hermes Agent, Devin, Goose, Cline, Kimi, Qwen Code, Amp, and Droid.

For example, you can have Claude Code handle one branch, Codex handle another, and an open-source agent like Oh My Pi run a cheap model for routine fixes, all in the same Orca window.

How to divide up work is suggested here (hypothetical example, not a recommendation from Stably). First, categorize the work:

  1. Hard work requiring long reasoning: assign to Anthropic's Claude Code or OpenAI's Codex if you have a subscription plan.
  2. Repetitive work like renaming variables or writing tests: assign to Oh My Pi with a cheap model.
  3. Work that needs comparison: send the same request to two agents then check the diff.

Each agent still logs in with your own account, so Orca doesn't charge extra fees.

How is Orca different from Cursor or running agents in a normal terminal?

Criterion Orca Cursor Terminal + tmux
Focus Orchestrating multiple CLI agents AI-powered code editor Managing sessions yourself
Agents Any CLI agent, using your subscriptions Cursor's agents, per plan Any CLI agent
Worktree Creates and tracks automatically Has background agent capabilities Do it yourself with git worktree
Reviewing changes Annotated diff, resend to agent Yes Use git yourself
Price Free, MIT Paid plans Free

Cursor works well when you want to write code yourself with AI assistance nearby. Orca works well when you've delegated most code writing to agents and need a place to manage multiple agents at once.

Installing Orca

Download the installer from onorca.dev/download (macOS Apple Silicon/Intel, Windows, Linux AppImage), or use a package manager:

# macOS (Homebrew)
brew install --cask stablyai/orca/orca

# Arch Linux (AUR)
yay -S stably-orca-bin

After installing, open the app, add a repo (your project folder with git), pick an agent you've already set up and logged into, then create your first worktree. The mobile app pairs with your computer to monitor from anywhere.

Controlling Orca from the command line

Orca has an orca command to let agents (or your scripts) control the app. Some commands, according to orca --help on version 1.4.229:

orca status                 # app and runtime status
orca repo add <path>        # add a project to Orca
orca worktree create        # create a new worktree managed by Orca
orca worktree ps            # summary of all worktrees
orca agent launch           # start an agent in a worktree
orca serve                  # run Orca headless on a server
orca automations create     # create scheduled tasks

There are also orchestration commands to let agents message each other, create and assign tasks, and set "decision gates" waiting for approval—useful when you want one agent to orchestrate other agents.

SSH setup in Orca: add a remote machine to run agents
SSH setup screen: add a remote machine as a place to run agents. Photo: official README.

What should I keep in mind before using it?

  • AI costs go up with more agents. Five agents running in parallel means five times the model usage. Monitor usage in Orca and set limits at your provider.
  • Disk space. Each worktree is a separate working directory; a Node.js project might need its libraries reinstalled in each worktree.
  • You still have to review. Orca helps you see diffs faster, but merging code to your main branch is still your call, especially when comparing results from multiple agents.
  • Usage data. Orca collects anonymous usage data and lets you turn it off; check the telemetry page before using it on sensitive projects.
  • Things change fast. The software updates almost daily, so the interface and commands might differ from this post. Check the GitHub changelog when something looks off.

Video review Orca

Some videos introducing and testing Orca (in English):

Christian Lempa: using Orca as a terminal and main editor.

Tonbi's AI Garage: installation, worktree and Design Mode.

Alex Hitt: git worktree and running agents in parallel.

AI Stack Engineer: running Claude Code, Codex and OpenCode at the same time.

Read more: Claude Code (or programming agent) for people who don't know how to code, 6 best tools for running multiple AI agents writing code at the same time in 2026.

Frequently asked questions

Is Orca free?
Yes. Orca is open source under the MIT license, according to GitHub (as of 19/10/2026). The cost only comes from the agents and AI models you use inside it, for example packages from Anthropic or OpenAI.

Does Orca write code by itself?
No. Orca orchestrates coding agents like Claude Code, Codex or Oh My Pi; the code writing is done by the agents.

I don't know git worktree, can I still use it?
Yes. Orca creates and manages worktrees for you. A basic understanding of git branches still helps when merging results from 2 or 3 agents.

Can Orca run on Windows?
Yes. Version 1.4.229 (17/10/2026) has builds for Windows, macOS and Linux, plus iOS and Android apps to monitor from remote.

Can I run agents on a server?
Yes, via SSH worktree or the orca serve command to run Orca without a UI on a remote machine. First add the remote machine in the SSH settings section, then create a worktree on that machine.

Conclusion

Orca is suitable if you already use coding agents daily and find waiting for one agent to finish sequentially to be a bottleneck. First, start small: 2 agents, 2 worktrees, one easy task, for example fixing a small bug or writing a test. Next, practice reviewing diffs and merging results with Git before increasing the number of agents, since the model costs from Anthropic, OpenAI or other providers increase per agent. Finally, check the changelog on GitHub whenever the UI or commands differ from this article, since version 1.4.229 (17/10/2026) may already be replaced by a newer version.

If you are new to coding agents, master one agent first. Read more in the post What is Oh My Pi? for more open source agent options.

Source: Orca README on GitHub, onorca.dev, orca --help command version 1.4.229. GitHub data retrieved on 19/10/2026.

Translated from Vietnamese with AI. Read the original.

Want to apply this in your business?

Book a free 30-minute call (Vietnam time, in English or Vietnamese).

Book a call