Blog Logo

Orca ADE: AI agents on multiple computers at once

Orca ADE is an open source development environment where the agents do the work and you direct them: terminals, files and orchestration in the same window. It solves the chaos of having OpenCode, Claude Code and Codex open at once, each in its own tab, with no idea who’s doing what, and it does it across several machines: your computer, a VPS and another local machine, with agents passing work between them. I’ve been using it for a couple of months now and, for the time being, I’m not going back.

What an ADE is and what it solves

An ADE is to an agent what an IDE is to a programmer: the place you open it from, configure it and watch what it does. Orca brings terminals, a file explorer, remote hosts and orchestration together in a single window, so you stop jumping between tabs. A new one comes out every week with buzzwords attached (skills, subagents, workflows), and with models that are already up to scratch, the problem has shifted to management: permissions scattered everywhere and contexts you copy and paste by hand.

Orca is open source and free, with the code in plain sight. If you know Warp, the interface will feel familiar, because it builds on the same idea of a modern terminal with workspaces. The difference is its agent focus: remote hosts, skills, automations and even a mobile app. Another example in this category is Antigravity, Google’s AI IDE, which I’ve already covered here.

How to set up a workspace with several agents

You set up a workspace in two steps: add a local folder and open as many tabs inside as you need, one per agent. Orca detects what you have installed and launches OpenCode, Claude Code or Codex in whichever tab you choose. In the settings you decide which agent is the default and whether it starts with full permissions or asks for confirmation before acting.

The steps, once you open Orca:

  1. Add a local folder as a workspace.
  2. An integrated terminal opens inside that workspace.
  3. Open new tabs: Orca suggests the available agents and launches the one you choose.

With several agents working in parallel on the same repo, you’ve got a real workflow. Watch out for your machine’s power: each tab takes its share, and opening five at once is noticeable.

Several computers at once with SSH hosts

SSH hosts let you have a VPS or a second machine as if it were local: it shows up in the sidebar with its files, its terminals and its own agents. You register it with its SSH credentials and open tabs on it without leaving Orca. The advantage is taking the load off your computer.

To add one:

  1. Open the settings and go to the hosts section.
  2. Add your VPS with its SSH credentials.
  3. Open terminals on that host and launch whatever you want: an OpenCode, a Claude Code, a deployment script.

The setup in the video is three machines: a local Mac, a VPS with the deployment environment and the agents’ always-on memory, and another local PC with more computing power. The heavy agents do the grunt work on the VPS while the laptop breathes.

Two details make the difference. Sessions run on tmux, so if the server restarts or the app updates, sessions stay alive and you pick them up where they were. And hosts come preconfigured: with a regular terminal you’d SSH in by hand every time; in Orca it’s just opening a tab and carrying on.

The orchestration skill: machines that pass work between them

Orca has an orchestration skill that installs with one click from the app itself. Once enabled, it coordinates terminals and different machines and passes data between them, so an agent on a VPS can ask another one running on your laptop for context.

The clearest case shows up in the video: an OpenCode running on the VPS communicates with the Claude Code in another tab on the same VPS and with the agents on the local PC. A Foundry session, a personal-use application, was created from another session running on the VPS. Information travels between terminals and between agents; you just tell it who it has to talk to and Orca handles the distribution.

It’s the same leap I already mentioned when talking about building a team of agents with MiniMax Code: going from a single agent that does everything to roles that collaborate. If you want the more technical version, I have a tutorial on agent orchestration in Go.

Files, automations and visible subagents

Orca shows you everything around each agent, all in the same window: the files of whichever host you have selected, the automations already running on each machine and the subagents in progress. That way you don’t have to open logs by hand to know what’s going on.

The file explorer changes depending on the selected host, and you can open and edit any file. A Markdown file shows up raw or with a preview, like on GitHub, and what you edit gets saved. It’s basically a mini editor with access to the code on all your machines.

The automations section detects the ones you already have on each host and brings them together in a single panel. On the video’s VPS there’s one that runs every day at 3 and checks for updates, created with Hermes, another of the available agents. If you manage a lot of workspaces, that panel is what keeps the chaos at bay.

Subagents are visible too. If you ask Claude to launch a research task with three subagents, the interface draws all three running on screen. You stop working blind.

When you open a repository you can set up scripts that run when you create a new worktree: install dependencies with npm install, copy config files, whatever you need. And in the accounts section you store your Codex and Anthropic accounts and see at a glance the limits you have left.

Local voice, computer use and a mobile app

Three extras come built in: voice dictation with a local model, computer control by an agent and a mobile app to follow sessions from your phone. Dictation saves you hiring an API, and computer use only worked locally in the tests; the app needs the machine running Orca to be on and connected.

The voice agent opens with a keystroke: you talk, it transcribes and answers, with a local model. There are more downloadable models, and only some of them use the OpenAI API.

The computer use skill gives it control of the computer, just like Claude’s computer use: it can browse, research websites or buy whatever you ask it to. On the VPS it wouldn’t start.

With the mobile app you control those same agents from your phone. You pair the phone once and your sessions are available on the sofa or away from home. It’s striking that this comes built into an open source app. Importing GitHub tasks caused problems in the video’s tests; the suspicion is a private organization GitHub.

When it’s worth it and when it’s overkill

Orca pays off if you work with several agents at once, if you want agents running on a VPS while your laptop rests, or if you manage several machines and need a single panel. With a single agent and a single repo, any terminal will do, and adding an ADE just adds noise.

Its limits, so you’re not caught off guard: each tab with agents consumes RAM and CPU, so opening loads of them willy-nilly ends up frying your machine. Computer use outside your local machine may not work. And plugins are still experimental.

For me, until something better comes along, it stays. Being open source, free and with the code in plain sight carries a lot of weight, and I don’t know of alternatives that combine remote hosts, orchestration and mobile while also being open source.

If you want to check it out, install Orca, open a folder as a workspace and launch two agents in separate tabs. If you already have a VPS, register it in hosts, install the orchestration skill and ask one agent to pass context to another.


What do you think?

Leave your opinion, question or suggestion. Comments are synced with GitHub Discussions .

← Back to blog