A desktop and mobile app to run Claude Code, Codex, OpenCode and other coding agents side by side, each in its own workspace, on your own subscriptions.
WhoDeveloper triaging a backlog who wants each ticket handled in a clean, separate checkout
The problemSwitching branches for every ticket leaves half-finished work behind and mixes agent changes together.
The outcomeEach task gets its own worktree and agent, and you review the result without leaving the app.
Steps
Connect GitHub or Linear in Orca following the review docs.
Pick a PR, issue or board item from the in-app list.
Open a worktree from that task and start your preferred agent in it.
Review the agent's changes in the diff view when it finishes.
Prompt for your agent
Using stablyai/orca, open a worktree from the GitHub issue or Linear task <issue link> and start Claude Code on it. Report which branch the worktree is on and verify it does not touch my main checkout.
Caveat. The README only says GitHub and Linear are supported natively, so other trackers such as Jira are not covered.
WhoDeveloper who kicks off long agent runs, then walks away from the desk
The problemYou keep checking a terminal to see if an agent is done or waiting on you.
The outcomeYour phone notifies you when an agent finishes, and you send the next instruction without going back to your computer.
Steps
Install the Orca desktop app and start an agent on a task.
Install the Orca mobile app from the iOS App Store or the Android APK link in the README.
Pair the phone with your desktop by following the mobile docs.
Walk away, wait for the finish notification, then send a follow-up from the phone.
Prompt for your agent
Using stablyai/orca, walk me through pairing the mobile companion app with my desktop Orca on <my OS>. Confirm what I should see on the phone when an agent finishes, and tell me how to verify notifications are enabled.
Caveat. The desktop app must stay running for the phone to reach your agents, and one issue thread suggests updating both phone and desktop if the companion misbehaves.
WhoDeveloper whose laptop is too slow or too small to run several agents and their builds at once
The problemParallel agents plus test runs overwhelm a laptop, but plain SSH sessions are fiddly to keep alive.
The outcomeAgents work in worktrees on the remote machine while you edit, review and use terminals locally, with reconnects handled for you.
Steps
Make sure you can SSH into the remote machine and that your agent CLI is installed there.
Add the remote host as an SSH worktree in Orca by following the SSH docs.
Start an agent in a remote worktree and watch it from the Orca window.
Use port forwarding to preview any dev server the agent starts.
Prompt for your agent
Using stablyai/orca, help me set up an SSH worktree for <user@host> and repo <path on remote>. Tell me how to confirm auto-reconnect works after I drop the connection.
Caveat. One HN user reported trouble installing on a headless server and fell back to a machine with a desktop, so a pure headless box may need extra effort.
WhoDeveloper stuck on a tricky refactor who wants more than one agent's attempt before committing
The problemTrying one agent at a time is slow, and agents sharing one checkout overwrite each other's changes.
The outcomeYou get several independent attempts at the same task, each on its own branch, and merge only the one you like.
Steps
Install Orca and open your repo in it.
Start Claude Code, Codex and OpenCode, each in its own worktree.
Paste the same prompt into each agent.
Review each worktree's diff side by side, then merge the winner.
Prompt for your agent
Using stablyai/orca, set up parallel worktrees in <my repo> and run Claude Code, Codex and OpenCode on this task: <task description>. Show me each agent's diff so I can compare, and verify no agent touched files outside its own worktree.
Caveat. Each agent burns tokens on your own subscription, so fanning one prompt across five agents multiplies usage and can hit rate limits fast.
Leave notes on specific diff lines and ship them to the agent so it fixes exactly what you flagged.
Beginner
15 minutes
From the README
Any agent
WhoReviewer who got a mostly right agent change and wants to correct a few lines without retyping context
The problemDescribing 'fix the third function in that file' in chat is vague and the agent often changes the wrong thing.
The outcomeThe agent receives line-level comments, revises the code, and you commit from the same window.
Steps
Open the agent's worktree and view its diff in Orca.
Click a changed line and write a comment saying what to fix.
Send the comments back to the agent.
Check the updated diff, then commit when it looks right.
Prompt for your agent
Using stablyai/orca, show me how to annotate lines in the AI diff for <worktree name> and send the comments back to the agent. Verify the agent's next diff addresses each comment.
Caveat. Annotations are only as good as the agent's willingness to follow them, so recheck the new diff instead of assuming every comment was applied.
Use Design Mode in the built-in browser to click any element and drop its code and a screenshot into your prompt.
Intermediate
30 minutes
From the README
Claude Code
Codex
WhoFront-end developer fixing a visual bug who cannot easily describe which element is wrong
The problemTelling an agent 'the button in the header looks off' leads to guesses about the wrong component.
The outcomeThe agent gets the element's real HTML, CSS and a cropped screenshot, so its fix targets the right code.
Steps
Run your web app and open it in Orca's embedded Chromium browser.
Turn on Design Mode and click the misbehaving element.
Send the captured HTML, CSS and screenshot into an agent prompt.
Ask the agent to fix the issue and check the result in the same browser.
Prompt for your agent
Using stablyai/orca, open <local dev URL> in Design Mode, capture the element <describe element>, and send it to Claude Code with the request: <what to change>. Verify the agent edited the component that renders that element.
Caveat. Design Mode captures what the browser renders, so framework-generated class names may not map obviously to source files and the agent may still need to search.
Use Computer Use so an agent can interact with visible desktop UI during a workflow.
Advanced
1 hour
From the README
Any agent
WhoDeveloper whose task requires poking at a native or visual app that has no API
The problemSome checks, like confirming a desktop app renders correctly, cannot be done from a terminal alone.
The outcomeThe agent operates the visible UI as part of the workflow and reports what it saw.
Steps
Read the Computer Use docs linked from the README.
Enable the feature for the agent you want to use.
Give the agent a narrow task on a specific app.
Watch its actions and stop it if it goes off track.
Prompt for your agent
Using stablyai/orca's Computer Use feature, have an agent open <app name> and perform <specific check>. Verify by reading back what the agent reports it clicked and saw.
Caveat. Computer Use lets an agent control visible apps on your machine, so run it only on non-sensitive sessions and watch what it clicks.
Use the Orca CLI so an agent can spin up worktrees and click through a UI on its own.
Advanced
1 hour
From the README
Any agent
WhoAdvanced builder automating a repeatable agent workflow instead of clicking through the app
The problemManually creating worktrees and prompting each agent does not scale when you run the same setup daily.
The outcomeA script or agent creates worktrees and drives the browser using Orca's own commands.
Steps
Read the Orca CLI overview docs to see the available commands.
Run orca worktree create for a new task branch.
Use the snapshot, click and fill commands to drive a page from a script or an agent.
Wire the commands into your own automation once one run works.
Prompt for your agent
Using stablyai/orca's CLI, write a short script that runs orca worktree create for <branch name> and then uses snapshot, click and fill on <local page URL>. Verify each command's output before chaining them.
Caveat. Only the command names worktree create, snapshot, click and fill appear in the README excerpt, so check the CLI docs for exact arguments before scripting.
Track rate-limit resets across your agent accounts and switch accounts without logging in again.
Beginner
15 minutes
From the README
Claude Code
Codex
WhoHeavy agent user who keeps hitting subscription limits in the middle of a task
The problemYou only learn a limit is hit when an agent stops, and re-logging into another account wastes time.
The outcomeYou see usage and reset times in one place and hot-swap to a second account to keep working.
Steps
Open Orca's account switcher and add your Claude or Codex accounts per the usage tracking docs.
Run your agents as usual and watch the usage readout.
When a limit is near, switch to the other account.
Resume the agent without a fresh login.
Prompt for your agent
Using stablyai/orca, explain how to add a second Claude or Codex account to the account switcher and read the usage and reset information. Verify which account an active agent is currently using.
Caveat. The README names Claude and Codex for usage tracking only, and swapping accounts to dodge limits may conflict with a provider's terms, so check your plan's rules.
Who it is forDevelopers who already pay for one or more coding agent CLIs (Claude Code, Codex, OpenCode and others) and want to run several at once without them stepping on each other's files.
Why it is hot+2,104 stars today, 98k total, picked up by one AI source we track
DEDUPED