Browser automation

Browser automation is useful for agents, scraping, UI tests, and computer-use workflows. It is also high-risk because browsers handle cookies, downloads, credentials, and untrusted web content.

Build a browser-capable image

Use a Nix flake target that includes the browser runtime and automation library:

cargo run -- build --flake . --profile browser

The build runs through the builder VM. The browser runs later inside the runtime microVM.

Run the automation

cargo run -- up browserbox --flake . --profile browser
cargo run -- exec browserbox -- node /work/automation.js

Security checklist

• Use a dedicated sandbox per browser session unless state reuse is intentional.
• Keep credentials out of the base image.
• Prefer short-lived secret references over copied cookie files.
• Restrict egress to the target domains when possible.
• Treat downloads as untrusted files.
• Snapshot browser state only when you are comfortable retaining cookies, cache, local storage, and downloaded content.

Planned runtime SDK shape

Status: Planned lifecycle API.

Python

browser = Sandbox.create(
    image="nix:./flake#browser",
    network=NetworkPolicy.deny_by_default().allow_https("example.com"),
)

try:
    browser.files.write("/work/automation.py", script.encode())
    result = browser.exec(["python", "/work/automation.py"])
finally:
    browser.stop()

TypeScript

const browser = await Sandbox.create({
  image: "nix:./flake#browser",
  network: NetworkPolicy.denyByDefault().allowHttps("example.com"),
});

try {
  await browser.files.write("/work/automation.js", script);
  const result = await browser.exec(["node", "/work/automation.js"]);
} finally {
  await browser.stop();
}