Skip to main content
Search public recall information for a sample make and model without using personal data.

Public-safety boundary

  • Blocker-aware research on a public safety site.
  • A hard privacy boundary: no real VINs or personal data.
  • Human-review routing before customer-facing safety decisions.

Setup

Use the CLI with a configured model profile. The example uses a sample make/model query and stays on public nhtsa.gov pages.

Run it from the CLI

web agent \
  --url https://www.nhtsa.gov/recalls \
  --allow-domain nhtsa.gov \
  "Search for public recall information for a sample 2020 Toyota Camry query. Return recall title, campaign number if visible, affected component, date, detail URL, and a short owner-facing summary. Do not enter a real VIN or personal data."

Expected output

2020 Toyota Camry recall information

1. Title: ...
   Campaign number: ...
   Component: ...
   Date: ...
   URL: ...
   Summary: ...

Build it into an app

const result = await agent.run({
  startUrl: "https://www.nhtsa.gov/recalls",
  goal:
    "Search for public recall information for a sample 2020 Toyota Camry query. Return recall title, campaign number, affected component, date, detail URL, and owner-facing summary. Do not enter a real VIN.",
});

if (result.status !== "completed") {
  await queueSafetyReview(result);
}

Production notes

  • Domain policy: nhtsa.gov.
  • Do not use a real VIN in public examples.
  • Include source URLs with any safety summary.
  • Add human review before using results in customer-facing safety decisions.

Inspect

  • Debug UI for form state and visible filters.
  • Steps for detail-page URLs.
  • Status for blockers or unavailable pages.
  • Recording for review.

Cleanup

For CLI-created browsers, cleanup is automatic. Close any browser you created separately while debugging with web browser close <browser-id>.

Common failures

  • If a flow asks for a real VIN, stop and switch to public make/model recall pages.
  • If recall data looks ambiguous, preserve the source URL and send the result to human review.