Dec 2025
Creating a run should be boring: one authenticated POST with a task description, a template id (if you are using a saved template), and an output schema. BrowserPilot returns a run id. That is it. No configuration, no magic, no side effects.
The request includes your task in plain language ('extract the first 10 products and their prices'), or references a saved template ('price-monitor'). It includes your output schema (a JSON shape for the result), and optionally any overrides to template parameters.
The response is fast and minimal: run id, status, created timestamp, and links to watch or poll. You do not have to wait for the run to finish. You kick it off and move on.
The observe path is where most agent APIs get awkward. Some require you to poll every few seconds to see if the run is done. Some only give you a URL to a dashboard. Some require you to hit a different endpoint for different data (one for status, one for result, one for logs).
BrowserPilot lets you choose: poll the run endpoint to check status and get the result, or register a webhook and we will notify you the moment the run reaches a terminal state (completed, failed, cancelled). The webhook includes the full result, so you often do not need to poll at all.
Everything the dashboard shows you about a run is available over the API: step history, screenshots, action log, cost breakdown, errors. There are no special admin endpoints or private API calls. If you can see it in the UI, you can get it via the API. An agent platform you cannot script is just a website.