Cross-platform computer-use agent achieving SOTA on desktop, web, and mobile benchmarks, surpassing human performance.
Surfer 2 is a cross-platform computer-use agent that runs on desktop, web, and mobile environments. It achieves state-of-the-art performance on major agent benchmarks (OSWorld, WebVoyager, WebArena, AndroidWorld), surpassing human baselines on desktop and mobile. The architecture separates strategic planning from tactical execution using an orchestrator module and sub-agents, following a ReAct loop with visual grounding, task validation, and failure recovery. Built by H Company, it integrates third-party frontier models and internal Holo1.5 models.
Key Features
check_circleCross-platform (desktop, web, mobile)
check_circleOrchestrator module for planning and coordination
check_circleSub-agents for tactical execution
check_circleReAct (reason+act) loop
check_circleVisual grounding
check_circleTask validation
check_circleFailure recovery
check_circleIntegration with third-party frontier models
lightbulbEnterprise IT teams automate complex desktop workflows across multiple applications, reducing manual effort by 80% and eliminating human error.
lightbulbWeb researchers use Surfer 2 to gather and synthesize information from hundreds of live websites, producing comprehensive reports in minutes instead of days.
lightbulbMobile app testers deploy Surfer 2 to automatically execute test scenarios on real Android devices, catching bugs before release and cutting QA cycles by 70%.
lightbulbData analysts configure Surfer 2 to extract and consolidate data from web dashboards and desktop tools, generating weekly business intelligence reports with zero manual intervention.
lightbulbCustomer support teams leverage Surfer 2 to navigate multiple CRM and ticketing systems simultaneously, resolving customer issues 3x faster by automating repetitive data entry and lookups.
lightbulbSoftware developers use Surfer 2 to automate cross-platform integration testing, ensuring their applications work seamlessly on desktop, web, and mobile without manual regression testing.
