Browser automation for people who hate CSS selectors

What it does

The repository is a thin Python client for Oxylabs’ Browser Agent, a hosted service that drives a real browser via API. You send a URL and a plain-English instruction—say, “find the price and take a screenshot”—and the cloud agent executes clicks, form fills, scrolling, and data extraction on your behalf. Results come back as JSON, Markdown, HTML, or base64-encoded PNGs, optionally shaped by an auto-generated or custom schema.

The interesting bit

The heavy lifting (vision parsing, DOM reasoning, action planning) happens entirely on Oxylabs’ servers; this repo is essentially a typed wrapper around their REST API. The real pitch is that you never write a CSS selector or sleep statement—the agent supposedly figures out the mechanics from your prompt, which is either liberating or opaque depending on your tolerance for black-box debugging.

Key highlights

• Natural language control over multi-step browsing sessions (click, type, navigate, wait, extract).
• Structured data extraction via OpenAPI-style schemas, with an optional schema generator.
• Four output formats: JSON, Markdown, HTML, and PNG screenshots.
• Geo-targeting support through proxy location parameters.
• Free trial includes 1,000 credits; paid plans start at $12/month for 3,000 credits and 1 req/s.

Caveats

• The README admits that “sites with advanced bot detection may require advanced setup,” so anti-bot success is not guaranteed out of the box.
• This is a client SDK, not a self-hosted engine; you are tethered to Oxylabs’ pricing, rate limits, and uptime.
• JSON output requires an explicit schema, which adds a configuration step the natural-language pitch tries to hide.

Verdict

Good fit for data teams that want to automate brittle web workflows without maintaining Selenium scripts, provided they are comfortable paying per request and trusting a remote black-box agent. If you need on-premise control, custom bot-evasion, or predictable latency, look elsewhere.