Browser automation and scraping at scale

Why it pays

• A real browser on a real desktop Chrome with a full profile, fonts and a GPU-free desktop, not a headless shim.
• Logins that persist Profiles live on the disk, so a logged-in session survives stop and resume.
• Fork a ready browser Prepare one logged-in browser, then fork it for parallel sessions.

Step by step

1. Start a sandbox and open its desktop

   Watch in your browser, or let an agent drive it.

   CLI

   ID=$(harebox new --name scraper --size small | jq -r .id)
   harebox desktop "$ID" --view browser

   Python

   box = hb.create(name="scraper", size={"preset": "small"})
   print(hb.create_desktop(box.id, body={"view": "browser"})["url"])

   TypeScript

   const box = await hb.create({ name: "scraper", size: { preset: "small" } });
   console.log((await hb.createDesktop(box.id, { body: { view: "browser" } })).url);

2. Log in once, then save the state

   Sign in by hand or with an agent, then save a template that keeps the profile.

   CLI

   harebox template save "$ID" --name logged-in

   Python

   hb.create_template(body={"name": "logged-in", "sandbox": box.id})

   TypeScript

   await hb.createTemplate({ body: { name: "logged-in", sandbox: box.id } });

3. Drive the browser with computer use

   Screenshot, click and type through the API, or let a harness use its built-in computer-use tools.

   CLI

   harebox computer "$ID" --action screenshot
   harebox computer "$ID" --action click --x 640 --y 360

   Python

   shot = hb.computer_action(box.id, body={"action": "screenshot"})
   hb.computer_action(box.id, body={"action": "click", "x": 640, "y": 360})

   TypeScript

   const shot = await hb.computerAction(box.id, { body: { action: "screenshot" } });
   await hb.computerAction(box.id, { body: { action: "click", x: 640, y: 360 } });

4. Record for audit

   Keep an MP4 of what the agent did.

   CLI

   harebox record start "$ID"

   Python

   rec = hb.create_recording(box.id)

   TypeScript

   const rec = await hb.createRecording(box.id);

5. Scale out from the template

   Start many browsers from the logged-in template, each with its own IP and disk.

   CLI

   harebox new --from logged-in --count 20 --label job=prices

   Python

   hb.create_sandbox_batch(body={"template": "logged-in", "count": 20, "labels": {"job": "prices"}})

   TypeScript

   await hb.createSandboxBatch({ body: { template: "logged-in", count: 20, labels: { job: "prices" } } });

Tips

• Respect each site's terms and robots rules. HareBox publishes abuse limits on egress.
• Keep a static IP on a sandbox when a site expects a stable address.
• Store scraped data in the sandbox and download it, or push it out by webhook when a run ends.