What Is Orca? Running Coding Agents Side by Side, Worktrees, Setup, Mobile App and Pricing

In Orca you open a task, and terminal coding agents like Claude Code, Codex and Pi run side by side on one screen. Each task opens in its own git working copy (worktree), so several agents can work on the same project without breaking each other's files; you compare the results and ship the one you like. It's free and open source (MIT), runs on macOS, Windows and Linux, and has a beta companion app for following along from your phone. Orca is not a model: the subscription or API fees of the agents you use are paid separately. Turkish can be selected as the interface language.

What it does: Runs coding agents side by side on one screen
What it runs: Claude Code, Codex, Pi and other terminal agents
Where: macOS, Windows, Linux · iOS and Android (beta)
Turkish interface: Yes (language option)
Agent cost: Separate: each agent's own plan
Permissions: By default agents start without asking for approval
Price: Free, MIT-licensed

What is Orca and what is it for?

Can I use it without coding?

No, Orca is not for people who don't code. The official documentation says so plainly: Orca is designed for people whose job is writing code and who use AI as leverage; if you're looking for a no-code tool, Orca isn't it. You're expected to be able to read changes (diffs) and work with git.

What problem does it solve? When you work with coding agents, running two agents in the same folder means they change each other's files and the work gets tangled. Waiting for one agent and then switching to the next is slow. Orca opens every task in a separate worktree: git creates a second, third working folder from the same project, and each folder has its own branch and its own files. So one agent fixes the login page while another works on the API, and you see the status of all of them in one window.

Run agents side by side without conflictsWhat you get: A separate branch and folder per task, status on one screenNeeds a separate setup Give the same job to three agents and compareWhat you get: Three branches and three diffs; the chosen one becomes a PR, the others are deleted with one clickNeeds a separate setup Annotate code line by line and send it to the agentWhat you get: A single fix prompt made of line-anchored notesBuilt into the app Watch agents from your phoneWhat you get: Status list, notifications, short repliesNeeds a separate setup

Orca is not a model; it runs the agents you already use (Claude Code, Codex, Pi, OpenCode…) and uses their subscriptions. Alongside the agents it adds terminal panes, a code editor, a built-in browser, diff review and GitHub/Linear task lists.

Who it's for: developers who run work with several agents at once, people who want several agents to tackle the same bug and then pick one result, and teams that want to run agents on a powerful remote machine and follow them from a phone.

Sources: Orca docs "What is Orca?" and "Worktrees", README (October 29, 2026).

How does Orca work? The worktree model

Task = worktreeFor each task, a separate folder and branch are created from the project's main branch.

You pick the agentA terminal opens in that folder; you pick the agent from the list and write your prompt.

Review, shipYou review the diff and leave notes; commit, push and PR happen inside the app.

A worktree is not a security sandbox: the agent can still reach any file and network its own process can access.

Each repo has a base branch (usually origin/main); new worktrees are created from it or from another branch or commit you choose.
Deleting a worktree deletes its folder and branch together (it asks for confirmation).
A new worktree is a clean copy: files git doesn't track, like .env, and folders like node_modules don't come along. You list files to copy in a .worktreeinclude file at the repo root, and large folders to link in the worktree.sharedDirectories list inside orca.yaml.
Creation runs in the background; meanwhile you keep working in other worktrees.

Source: Orca docs "Worktrees" and "orca.yaml & .worktreeinclude" (October 29, 2026).

How do you download and install Orca? macOS, Windows, Linux, phone

The official download address is onorca.dev/download; the same files are on the GitHub Releases page. Orca launches the agent and uses the agent's own account: the agent itself must be installed and signed in. Orca can install most agents on the supported list with one click from the agent picker; if an agent won't start, first run its command manually in a terminal.

orca-macos-arm64.dmg for Apple Silicon, orca-macos-x64.dmg for Intel. The app is signed and notarized by Apple; on first launch macOS may ask for approval, which is normal.

With Homebrew› brew install --cask stablyai/orca/orca

Always type the full name: a bare orca on Homebrew is an unrelated package. To update: brew upgrade --cask stablyai/orca/orca

Download and run the orca-windows-setup.exe installer. The default shell is chosen under Settings then Terminal, PowerShell or CMD; the docs recommend PowerShell for most users.

AppImage: Available for x64 and arm64; needs chmod +x before the first run. It updates itself.
.deb / .rpm: Installed with the package manager; when there's a new version, Orca asks you to copy the install command and doesn't update itself.
Command name: On Linux the Orca CLI ships as orca-ide; the name orca belongs to the GNOME screen reader.

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

What happens on first launch?

It asks for access to your home folder so you can add repos.
If present, it offers to import your ~/.claude, ~/.codex and Ghostty terminal settings.
On the empty start screen you add your first repo (Add Repo).

Do you need an account? The desktop app doesn't need an Orca account. If you pair your phone through Orca Relay, you're asked to sign in. Updates: by default it updates itself from the stable channel. The latest version is 1.4.243 (October 28, 2026). The official documentation doesn't state a minimum operating system version or RAM.

Sources: Orca docs "Install", "Mobile companion", README, GitHub Releases, App Store listing (October 29, 2026).

How do you use Orca? What you type, what you get

Add the repo. Use Add Repo in the sidebar to pick a project folder on your computer.
Open a worktree. Click the + next to the repo name and type a task name (e.g. login-bug). If you leave it blank, Orca names it after a sea creature.
Pick the agent. In the terminal that opens, pick Claude Code, Codex or another agent from the list and write your prompt.
Watch and review. The sidebar shows whether each agent is working, done or waiting for you; you switch between worktrees with Cmd+J (macOS) / Ctrl+Shift+J (Windows, Linux).

Example 1: Have three agents fix the same bug"After logging in, users are sometimes sent back to the login screen. Find the cause, fix it, and add a test that catches this case."What you do: open three worktrees from the same base branch (login-bug, login-bug-2, login-bug-3), pick a different agent in each and paste this prompt. What you get: per the official description, three separate branches and three diffs; you review the one you choose and send it as a PR, and delete the other two with one click. The docs explain the logic like this: where the agents agree is probably right, and where they diverge is the hard part of the job.

Example 2: Send the code review back to the agent"This function throws an error when it gets an empty list; return 0 for an empty list."What you do: in the worktree's diff view, click the + next to the relevant line and write this note, add notes to other lines too, then press Send to agent and pick the agent. What you get: Orca turns all the notes, with their line numbers, into a single prompt and sends it to the agent; the agent fixes everything in one go. Notes stay on the line after the fix; you check the fix and close them with Resolve.

Example 3: Point to an element on the page to get it fixed"This button overflows on mobile; make it fit without wrapping the text onto two lines."What you do: in the worktree's built-in browser, click the button on the page with Design Mode. What you get: according to the README, Orca adds that element's HTML, CSS and a cropped screenshot to the agent's prompt; the agent doesn't have to hunt for which element you mean.

From the command line: the Orca CLI controls the app from a script or from the agent itself. It's registered under Settings then General then Orca CLI. Examples from the docs:

Open a worktree from an issue, send a reply to the terminal

› orca worktree create --repo id:<repoId> --name my-task --issue 123 --json
orca terminal send --text "continue" --enter --json

Running agents in parallel uses up each agent's own usage limit. Before putting three agents on the same job, check your plans' limits; Orca shows usage and reset time for Claude and Codex.

Sources: Orca docs "Your first 3-agent session", "Race three agents on the same task", "Annotate AI Diff", "Orca CLI overview", README (Design Mode) (October 29, 2026). Outputs are described as in the official documentation.

Which agents does Orca work with? Claude Code, Codex, Pi…

The README's rule is simple: any coding agent that runs in a terminal also runs in Orca. Some come preset in the list (one-click launch and install); some have deeper integration such as status tracking, usage limits and account switching. The home page says "27 supported agents"; the table in the docs is longer and keeps growing.

Agent	Integration in Orca (per the docs)
Claude Code	Deep: usage tracking, account switching, hooks
Codex	Deep: usage tracking, account switching
Cursor CLI	Deep integration
Pi	Auto-install, hooks, status
OpenCode	Auto-install, status
Gemini CLI, GitHub Copilot CLI, Grok, Amp, Cline, Goose, Kimi, Qwen Code…	Auto-install

The agent's cost is separate

Orca doesn't offer its own model; it launches the agent with your account. For Claude Code your Claude plan applies, for Codex your ChatGPT plan or your API fees.

Usage and accounts

It shows usage and limit reset time for Claude and Codex. It also lets you keep several Claude accounts saved and switch between them; for the terms on using a Claude subscription in third-party tools, see Anthropic's Claude Code legal page.

If an agent won't start

Per the docs, first run the agent's command manually in a terminal; if it errors there too, the problem isn't Orca but the agent's installation or sign-in.

Sources: README "Supported Agents", Orca docs "Supported agents" and "Troubleshooting", home page (October 29, 2026). The table is an excerpt of the official list; the full list is in the docs.

How do you use Orca from your phone and on a remote machine?

Desktop appAdding a repo and opening a separate worktree for each task, running agents side by side, reviewing and shipping changes.Desktop · Free; macOS, Windows, Linux; requires the agent's own account · details

Phone companion app (beta)Watching desktop agents from your phone, replying to waiting agents, getting notifications.Mobile · Beta; iOS and Android; pairs with the desktop · details

SSH and remote Orca serverRunning agents on a powerful remote machine or your own always-on server.Desktop, Terminal · The machine or cloud account is yours; Orca doesn't sell hosting · details

Orca CLIControlling Orca from the command line or from the agent itself.Terminal · Ships with the app; registered in Settings; called orca-ide on Linux · details

Phone companion app (beta)

The phone app isn't a full editor; it's a remote control for Orca on your desktop, and the desktop still does the work. What you can do:

See each worktree's agent and status (working, done, waiting for you); get a notification when an agent finishes.
Write a short reply to a waiting agent (continue, yes or free text), attach a photo or dictate by voice.
Look at changed files, commit; switch the agent account and see usage.

On the desktop, open the pairing screen; a one-time code appears.
On the phone, tap Pair and paste the code. Pairing through Orca Relay requires signing in; you can also connect by entering an address on the same local network or via Tailscale.

If the desktop app closes, a direct or local-network connection drops; when you open the desktop again, the phone reconnects. The code expires within a few minutes.

Agents on a remote machine over SSH

For long builds or work that needs a powerful machine, you add a server under Settings then SSH; when creating a worktree you choose that server instead of Local. The worktree and agent run remotely, while the editor and diff open on your laptop as if they were local. If the connection drops, the agent keeps going; Orca reconnects and brings the terminal back.

On the first connection, Orca installs a small component of its own on the remote machine. If a Linux server lacks make, a C++ compiler and python3, files and git work but the remote terminal won't open.

Always-on server: you can run Orca as a server on a Mac mini at home or on your own VPS and connect your laptop, browser and phone to it; agents keep working while your laptop sleeps. Orca doesn't sell hosting: the machine and the cloud account are yours.

Sources: Orca docs "Mobile companion", "SSH worktrees", "Ways to run Orca" (October 29, 2026).

Is Orca free? Pricing and license

Yes. Official FAQ: Orca is completely free and open source under the MIT license, and downloads free on macOS, Windows and Linux. There's no paid individual plan; onorca.dev has no pricing page. The phone app is listed as free on the App Store.

Item	Status
Orca desktop and phone app	Free (MIT)
The agents you use	Separate: each agent's own subscription or API fees
Remote machine, VPS, cloud environment	From your own account; Orca doesn't sell it
Enterprise	Enterprise sales contact available; no price on the official page

Allowed

Using it for commercial work, copying, modifying, distributing (MIT)

Not allowed

Removing the copyright and license notice from copies
The software is provided "as is"; the developer gives no warranty

Sources: onorca.dev FAQ, README "License", Enterprise page, App Store (October 29, 2026). This is not legal advice; the terms of the agents you use also apply.

Official pricing page last checked on November 3, 2026.

Does Orca work in Turkish?

Desktop interfaceYes	The Language option in the Appearance section of the settings includes Turkish; if System is selected, it follows the operating system's language.
Agent repliesUnclear	Orca has no replies of its own; replies in Turkish depend on the agent and model you use.
DocumentationPartly	The README has a Turkish translation (docs/readme/README.tr.md); the docs site is in English.
Phone appUnclear	The App Store listing shows English only; a Turkish interface isn't clear from official sources.

Sources: Orca docs "Settings", GitHub repository, App Store listing (October 29, 2026).

Is Orca safe? Permissions, my code and my data

By default Orca launches agents with flags that let them run without asking for approval (such as --dangerously-skip-permissions for Claude); if you want to approve every command, choose Manual under Settings then Agents then Agent Permissions.

Where your code goes

According to the official Enterprise page, prompts and code go directly to the agent provider you configured (Anthropic, OpenAI…); Orca doesn't inspect or store them. The data rules are those of the agent you use.

Anonymous usage data

Orca collects anonymous usage data (PostHog, USA): version, operating system, which agents you use and token counts. File contents, prompts, agent output, repo and branch names are not sent. To turn it off: Settings then Privacy then Share anonymous usage data or DO_NOT_TRACK=1.

A worktree is not a sandbox

A worktree is a separate folder, not a security box: the agent can reach every file and network its own process can access.

Access on first launch

It asks for access to your home folder so you can add repos; it offers to import your ~/.claude and ~/.codex settings.

Sources: Orca docs "Supported agents" (Permission safety), "Privacy & Telemetry", "Install"; Enterprise page (October 29, 2026). This is not a security audit.

Common Orca problems and fixes

What happens	What to do
The agent won't start	Run the agent's command manually in a terminal; if it errors there too, the problem is the agent's installation or sign-in. Check in Settings then Agents that the command is on the PATH.
A worktree can't be created	The starting branch may not have been fetched: run `git fetch origin` in the repo. If a worktree already exists for that branch, delete it or use a new name.
The new worktree has no .env or node_modules	A worktree is a clean copy. List files to copy in `.worktreeinclude`, and folders to share in `worktree.sharedDirectories` inside `orca.yaml`.
"orca: command not found"	Register it under Settings then General then Orca CLI; on macOS, ~/.local/bin must be on the PATH. On Linux the command is `orca-ide`.
The phone won't pair	Desktop and phone must be on the same Orca account; the code expires within a few minutes, so generate a new one.
SSH connects but the remote terminal won't open	Install make, a C++ compiler and python3 on the Linux server, then reconnect.
The computer slowed down, memory is full	Close worktrees you aren't using (each one keeps a file watcher); split layouts with many browser tabs use the most memory.

Source: Orca docs "Troubleshooting & FAQ", "Mobile companion", "Worktrees" (October 29, 2026). For error logs: Help › "Open Logs".

Does Orca replace Cursor or Claude Code?

Orca's own FAQ describes it as "a terminal-first Cursor alternative": instead of a classic code editor, an environment that manages several agents and worktrees. But it doesn't replace Claude Code or Codex; it runs them. So you still install the agent and pay for it, and Orca lets you run them in an organized, parallel way.

The rows for other tools summarize those tools' own guides.

Frequently asked questions

What is Orca and what is it for?

A free, open-source desktop app that runs terminal coding agents like Claude Code, Codex and Pi side by side on one screen, each in its own git worktree. The agents don't break each other's files; you compare the results and ship the one you like.

What is a worktree in Orca?

It's a separate working folder git creates from the same project, with its own branch and files. Orca opens every task in such a folder; deleting it removes the folder and branch together.

Is Orca free?

Yes, it's MIT-licensed open source and free. However, the subscription or API fees of the agents you use (Claude Code, Codex, etc.) are paid separately; Orca doesn't offer its own model.

Is Orca a Cursor alternative?

Orca's own FAQ describes it as a terminal-first Cursor alternative. Instead of a classic editor, it's an environment that manages several agents; Cursor CLI can also run in Orca as an agent.

Can you use Orca from your phone?

Yes, there's a beta companion app for iOS and Android. It pairs with Orca on your desktop; you see the agents' status, get notifications and reply to waiting agents. The desktop does the work.

Does Orca work on Windows?

Yes. There's a Windows installer; there are also macOS (Apple Silicon and Intel) and Linux (AppImage, deb, rpm) versions.

Does Orca work with the Pi coding agent?

Yes. Pi is on Orca's preset agent list, with auto-install, hook and status tracking support.

Related guides

Sources

Not clear in the official sources: Minimum operating system version and RAM for desktop · the phone app's Turkish interface · number of supported agents (the site says 27, the list in the docs is longer) · Enterprise pricing · steps to uninstall the app and delete settings. Each is covered in the relevant section along with what you can do.