Short answer: Orca is a desktop app that runs AI coding agents like Claude Code and Codex side by side, organized by project. I used it for about a week. If you pay for one AI plan, you can skip it: that company's own app will work better on its own. If you pay for more than one, like I do with Claude Max and ChatGPT Pro, Orca saves a lot of jumping between apps. My first impressions are great, and I'm going to keep using it.
I use AI agents all day: ad creative tests, outreach, lead lists, and a stack of scheduled routines that run every morning and night without me having to get involved. Before Orca, that work was spread across the Claude app, the Codex app, my own agent setup and Telegram, on more than one computer. Here's what changed when I moved it into one window.
What Orca is
Orca is a free, open-source desktop app for Windows, macOS, and Linux. It runs coding agents such as Claude and ChatGPT in parallel, each in its own git worktree, so two agents working in the same project don't trip over each other's changes. The agents run on the Claude or ChatGPT subscription you already have.
In practice, each of your projects sits in a list on the left. Open one and you get a terminal with your default agent. Mine is Claude. You can open a Codex terminal in the same project and run both side by side.
Who should skip it
If you only pay for one AI subscription, I wouldn't bother with Orca yet. Claude's app is built around Claude, and ChatGPT's around ChatGPT. When everything lives with one company, its own app will usually work better.
Orca makes sense when you pay for more than one, which a lot of people now do. That's where the time goes: switching apps, re-explaining context or performing hand-offs, and thinking about whether the right agent has the right access, connectors and setup.
What works
Everything for a project in one place
The biggest win is not having to ask, "Is this project set up the same on this laptop? Are the same files and tools connected?" The project, its files and its agents are in one window. That sounds small until you count how often you switch apps in a day.
Claude directing ChatGPT agents
You can have Claude direct Codex agents, and it works in either direction.
In the video, I asked Claude Opus 5.5 to spin up three Codex agents to research three pickleball players for a site I run, PaddleSetups. Claude wrote each ChatGPT agent a research brief. Each one started in its own Orca worktree, and Claude laid out the rules: every fact needs a dated source, anything unproven gets marked "Pending," and nothing gets pushed or published without my approval. I could watch all three work at once.
All my automations in one list
If you run AI automations, you know the questions. Did it run? Did it hang? What did it deliver? Orca puts my live automations in one list, with the agent each one uses, the project it belongs to and when it last ran. If you are living and reading this in the month of October 2026, you'll see most of my automations are set to run on Claude Opus 5.5, with a few on ChatGPT/Codex to spread out usage.
The bigger benefit shows up when an automation needs a change. Before, I had to remember whether it lived in Claude or in Codex, then find the right instruction file. Now I can see them all and make the change in one place.
Voice typing
Orca's voice typing (Ctrl+E by default) is just short of being on par with Claude's and, in my opinion, better than ChatGPT's. Codex shows a sound wave while you talk. Orca, while not perfect, shows your words in chunks as you say them, so you can mostly tell when it's keeping up. Claude has them both beat when you have a solid internet connection, since it shows the words live as you say them.
A mobile app
Orca's mobile app looks a lot like the desktop version and has nearly all the same features, so you can check on agents away from your desk.
What doesn't work yet
- Split layouts don't follow you between projects. I'd like one company's project in one pane, another company's in a second and something personal in a third. When I switch projects, Orca goes back to that project's own layout. There may be a workaround I haven't found.
- Voice typing can lose track when you leave the app. It's reliable inside Orca. When I jumped between other tabs and browsers, it didn't always keep up.
- No agent that talks back. I'd love to walk around and have a spoken conversation with an agent that knows what's going on. That isn't there yet. I've even connected Claude Duet to Orca, but it's more effective to just deliver projects to your PC through Hermes on Telegram or another method, since Claude Duet wouldn't necessarily be able to access your PC at this time and give you progress on these agents. Maybe I could have Orca hand a task back to Duet through a calendar notification or email when it finishes or makes progress? 🤔 Honestly, that's probably overkill. Given the current stage of things and how fast they're progressing, it's better to just use one app.
My Orca first impressions: the verdict
First impressions are great, and I'm going to keep using it. If you use one platform and don't need to pick between models, stick with Claude or ChatGPT on their own (or Hermes, Grok Bot, Open Bot, whatever your current preference is). If you pay for more than one subscription, Orca is the move right now.
What this means for a small business
You don't need to write code to get something out of this. The same idea applies to any business running AI agents: managing lead lists, follow-ups, morning reports. Problems start when those agents are scattered across apps and nobody can see what ran. Try keeping your AI work in one place (ideally while staying model-flexible), and decide up front which messages need a human in the loop or a person's approval before they go out.
That's how we build client systems at My Robot Worker, whether it's an AI Front Desk that answers calls and texts or an agent that follows up on new leads. The rules get tested before anything goes live. Prices are on the pricing page.
Frequently asked questions
Can Claude Code and Codex work together?
Yes. In Orca, I had Claude Opus 5.5 hand research tasks to three Codex agents at once, each working in its own copy of the project. It works the other way around too, with Codex directing Claude agents.
Do I need a separate AI subscription for Orca?
No. Orca itself is free and open source, and the agents run on your own subscription(s).
Does Orca work on Windows?
Yes. I use it on Windows, and Orca's GitHub page lists macOS and Linux too.
Is Orca worth it if I only use ChatGPT?
Probably not yet. If all your AI work runs through one company, its own app is the simpler choice.
Not sure where AI agents fit in your business?
The AI Opportunity Consult is a 60-minute, $250 working session. We look at the repetitive work in your business and map out where an AI agent would actually help. If you move forward with a plan, the full $250 is credited toward your setup.