Browser Use Orchestrator icon

Browser Use Orchestrator

Pilote browser-use-cdp-orchestrator (agents web, sessions, jobs, profils)

Actions16

Agent → Run Vision Agent

AI-generated

Summary

Captures a full-page screenshot of a URL through the browser-use orchestrator and returns a vision-LLM description of it. It can also run an agent first to interact with the page before the capture.

Inputs

  • task (required) — The URL to describe in direct mode. If Direct is false, a natural-language instruction the agent runs before the capture.
  • additionalFields.direct — The server-side default is true: navigate straight to the URL, then capture and describe with no agent loop (about 10–30s). Set false to have a browser-use agent carry out the task step by step first (slower, can interact). The n8n checkbox defaults to false, and the node only sends the field if you add it.
  • additionalFields.screenshot — Saves the full-page JPEG in screenshots/ and returns its HTTP URL in the result's url field. The server-side default is true.
  • additionalFields.model — LLM model ID for this job only. If empty, the server's .env models are used.
  • additionalFields.profile_id — Browser-use profile ID to load, which keeps cookies and logins. If omitted, the browser starts clean.
  • additionalFields.proxy_country — Two-letter proxy country code, such as fr, us or de, for geo-restricted content.
  • additionalFields.wait_seconds — Maximum synchronous wait on the server, 0–600. 0 launches the job without waiting.
  • executionMode — Sync (default) waits for the job to finish by polling Await Job internally. Async returns a job_id immediately, and the node forces wait_seconds to 0.
  • syncMaxWaitSeconds — Sync mode only. Total maximum wait, 300s by default. If the job isn't finished by then, the node returns the current state including the job_id, and the job keeps running on the server.

Output shape

One item per input item. It holds the job result JSON from the orchestrator, which includes the vision description and, if screenshot is enabled, a url for the image. In async mode, or if sync times out, it holds the job state with a job_id.

The job status is done, error or cancelled once it finishes. The exact result field names beyond url aren't defined in the node source. With Continue On Fail enabled, errors come back as {error: message}.

Examples

Example 1: Describe a web page visually

task = https://news.ycombinator.com; leave Direct at its default; executionMode = sync.

Example 2: Log in, then capture and describe the page

task = natural-language instructions; Additional Fields: direct = false, profile_id = saved profile.

Example 3: Start a long capture and collect it later

executionMode = async; store job_id; later use Job > Await Job.

Links

Discussion