Owl Browser

@fhsethen Hello! Congratulations on the launch. I browsed quickly your homepage and I couldn’t see any trial or free plan, the self-hosted seems to be even more expensive than the starting cloud subscription. Who is your target with this product and do you plan a way to try it without committing large sums?

@fhsethen  @amberjolie Thanks for the support! You hit on a massive pain point—modern web apps are incredibly fluid, and layout shifts are the enemy of traditional automation.

To answer your question on Dynamic UIs: We tackle this with a two-layer approach:

1. Semantic Targeting: Instead of brittle CSS selectors, we use Natural Language Selectors. Our engine uses a multi-scorer ensemble (analyzing text similarity, visual proximity, and element type) to locate the correct element conceptually, even if the DOM attributes have shifted.

2. Execution Stability: Once we identify the target, we employ a "verify-and-retry" mechanism for the interaction itself. Before a click is finalized, the system calculates the precise element position and performs a rapid stability check (a 10ms micro-wait). If the UI shifts during that window (e.g., a loading spinner expands), we detect that the action didn't land effectively via our Action Result feedback loop. The system automatically retries this process up to 10 times to catch the element in a stable state.

Regarding Safety and Loops: That same retry logic acts as our circuit breaker. Because every action is part of a parsed "action plan" generated by our Natural Language Automation (NLA) engine, we don't just loop blindly. If the 10-try threshold is reached without a confirmed success, the specific action fails, triggering our error recovery protocols rather than spiraling into an infinite clicking loop.

It’s all about balancing agent autonomy with strict execution boundaries! 🦉 

@mykyta_semenov_ for this, you will need residential IPs. Our browser can scale like no other browser. You can spawn 100 contexts with 100 different proxies and scrape 100 pages in less than a minute (if not login required).

@ahkeminozen This is a browser as a service and we currently don't support any extensions. We tried to build all power browser managements as tools.

@yamamoto7 The browser uses our custom DOM, which allows us to find any element even on heavily JavaScript-heavy pages. We employ three different strategies for finding elements: the Semantic Matcher, LLM-based Matcher, and Vision Matcher. While the LLM and Vision matchers can be slower, they serve as a last resort; 85% of elements are matched using the Semantic Matcher, which utilizes our custom-trained ML model.