Isolade
Docs

Quickstart

From install to a running agent in about five minutes.

This walks through installing Isolade and getting an agent working in its first profile.

1. Install

One command covers both platforms. It detects your OS and architecture and runs the right steps:

curl -fsSL https://isolade.com/install.sh | sh

On Linux you’ll need KVM and membership in the kvm group. See Installation for the full requirements.

2. Follow the guided setup

A fresh install has no profile, and a profile is what an agent runs in, so the app opens onto guided setup. It takes four steps and leaves you with an ordinary profile, built and signed in. You can reopen it any time from Settings → Profiles.

Choose what to work on. Either the Excalidraw demo, which is a real project set up for you, or your own code. Pick the demo if you want to watch an agent work before configuring anything.

Review the Dockerfile. Guided setup writes it from your answers: a base image, the toolchains you ticked, and one COPY per repository. It is a normal file, editable here and afterwards under Settings → Dockerfile. A scaffold deliberately installs no project dependencies, so the first build cannot fail on them. Ask the agent to install them in the VM, and add the line you needed once you know it.

Build. The first build of an install is the slow one: BuildKit boots in its own microVM and the base image is pulled, and both are cached for every build after. Expect minutes, watch the log, and note that closing the card does not stop the build.

Sign in. Claude or Codex, whichever you have. The login runs the provider’s own CLI inside a throwaway VM, your browser opens the authorize page, and the callback finishes it. There are no codes to copy, and no Isolade account. This step comes last because the login VM boots from the image you just built. Afterwards the same logins live under Settings → Providers, and the model picker offers whichever of the two you have signed in to.

Isolade signs in like a second device, so the claude and codex CLIs on your machine keep their own logins. See Agent accounts for the details.

3. Run an agent

Start a chat. Each one gets a fresh microVM built from the profile. Pick a model, give the agent a task, and it works unattended inside the VM. Watch the diff count climb in the sidebar, and review the changes when it’s done.

If you took the demo, “start the dev server” is a good first task: the port is already forwarded, so Excalidraw appears in the Browser tab beside the chat.

Doing it by hand

Guided setup is a shortcut, not a special path. The same profile can be authored directly:

  1. Settings → Profiles, create one and name it.

  2. Configuration, point it at a repo (a local checkout or a GitHub URL).

  3. Dockerfile, write it. Each repo reaches the build as a named context, keyed by its name, that the Dockerfile copies from:

    FROM ubuntu:24.04
WORKDIR /workspace/myapp
COPY --from=myapp . .

    Isolade renames your last stage and stacks the agent tooling on top, so you don’t install claude or codex yourself.

  4. Build, and press Build.

Next steps

  • Add a secret so the agent can push to GitHub without ever seeing your token.
  • Spread the profile across several repos.
  • Mount a build cache so rebuilds stay fast.