Browser Use Orchestrator icon

Browser Use Orchestrator

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

Actions16

Agent → Run Task Agent

AI-generated

Summary

Launches an autonomous browser-use agent that carries out a natural-language web task (navigate, click, fill, extract) through the orchestrator's REST launch endpoint. It returns the job result, or a job_id if the job is still running.

Inputs

  • task (required) — Natural-language instruction. Include the start URL and the expected result to avoid needless verification steps.
  • additionalFields.vision — Send per-step screenshots to the LLM. Slower and costlier. Default false (DOM only).
  • additionalFields.screenshot — Capture the final page as a full-page JPEG and add its URL in result.screenshot_url.
  • additionalFields.model — LLM model ID for this job. Empty uses the server's OPENAI_MODEL.
  • additionalFields.fields — JSON schema of the results as { field_name: description }. Max 12 identifier-named fields. A type prefix such as number:, bool: or int: is optional. When omitted, results is free text.
  • additionalFields.secrets — Names of server-side secrets made available to the agent. The values never pass through the LLM. Refer to them by name in the task.
  • additionalFields.allowed_domains — Navigation allow-list (wildcards such as *.google.com are allowed). Navigation outside the list is blocked. Recommended when secrets are used.
  • additionalFields.profile_id — Browser profile ID to load, which keeps cookies and logins. Empty starts a blank browser.
  • additionalFields.proxy_country — Two-letter proxy country code, for example fr or us.
  • additionalFields.wait_seconds — Server-side wait for the launch call, 0–600. 0 launches without waiting.
  • executionMode — Synchronous (default) launches the job, then repeatedly calls Await Job until it finishes or Max Wait expires. Asynchronous returns a job_id immediately, and the node forces wait_seconds to 0.
  • syncMaxWaitSeconds — Synchronous mode only. Total maximum wait in seconds, default 300. If the job is not finished, the current state with its job_id is returned and the job keeps running server-side.

Output shape

One item per input item, whose JSON is the job state or result from the orchestrator. It contains a status (done, error or cancelled once finished), a job_id, and the result. With fields set, results holds objects with those keys. With screenshot enabled, result.screenshot_url is included.

In async mode, or when Max Wait runs out, the output is a non-terminal state with a job_id. Pass that job_id to Await Job or Cancel Job. The exact result shape comes from the server and is not defined in the node. With Continue On Fail enabled, errors are returned as { error: message }. The node needs the browserUseOrchestratorApi credential (base URL).

Examples

Example 1: Extract the top Hacker News posts as structured data

task: "Go to news.ycombinator.com and give the first 3 posts"; fields: {"title":"post title","url":"post url"}; Execution Mode: Synchronous

Example 2: Log in to a site with a stored secret, restricted to one domain

task referencing the secret names; secrets: [my_login, my_password]; allowed_domains: [example.com]; profile_id set to keep the session

Example 3: Start a long task and collect it later

Execution Mode: Asynchronous; store the returned job_id; follow with Job > Await Job

Links

Discussion