H Company Debuts Runner H Public Beta, Open-Sources Surfer H Visual Model and Launches Tester H Platform

Agentic AI theory holds that complex projects can be broken into coordinated steps carried out by multiple specialized agents. That concept has lingered on the horizon rather than delivered hands-on value. Paris-based H Company aims to change that with three key launches that establish a real-world route for multi-agent workflows across engineering and business tasks.

At the heart of this initiative is Runner H, which has entered public beta to let a broad audience experiment with end-to-end AI-driven automation. A single high-level prompt sets in motion an entire team of sub-agents that together handle everything from data collection and document creation to messaging and transactional processing. Alongside this public rollout, H Company open-sourced the vision-language model powering its browser automation module, Surfer H, and introduced Tester H in private beta as an AI-based testing engine for enterprise applications. These three releases form a cohesive toolkit for automating tasks that previously required stitching together scripts, integration platforms and manual checks.

Runner H itself sits on a modular architecture composed of four primary layers: memory management, task planning, action execution and monitoring. The memory layer converts user inputs, document contents and extracted data into structured vectors, ensuring context persists across multiple requests. In the planning layer, a directed acyclic graph breaks down a submitted goal into discrete subtasks, assigning each one to an appropriate sub-agent. During execution, those sub-agents may call APIs, simulate browser interactions or transform data formats. A monitoring component captures success metrics, logs and exception reports in real time. This design addresses two persistent drawbacks in language-model automation—loss of context across steps and erratic task completion—by verifying each action and maintaining a continuous state throughout the workflow.

In practice, Runner H can handle a wide variety of scenarios:

• Automated research and reporting: Crawl multiple websites or public databases, extract relevant details, assemble a structured summary and distribute the final report via a communication channel.
• Job application management: Search career portals for openings that match predefined criteria, customize application materials and submit each application automatically.
• CRM enrichment: Pull contact details from emails or social platforms, update records in a CRM system and draft custom follow-up messages.
• Invoice processing and payment: Download vendor invoices, match them to purchase orders, prepare payment requests for approval and execute settlements once authorized.
• Travel and event coordination: Compare flight and hotel options within policy guidelines, generate itineraries and share confirmations with participants.

Runner H also integrates with leading productivity and data tools, including Slack, Notion, Google Sheets and Microsoft Teams, alongside an expanding set of RESTful connectors. Uploaded PDFs, Word documents, CSV files or entire content repositories are parsed, indexed and added to the system’s working memory. A single interface lets teams inspect each step, tweak agent parameters and review detailed logs of every API call or user-interface interaction. For security-sensitive environments, Runner H can run in private clouds or on-premise clusters with support for single sign-on, role-based access control and end-to-end encryption.

Imagine a product manager who needs a daily competitor pricing analysis. After granting access, she instructs Runner H to fetch prices for a set of SKU codes from multiple e-commerce sites, normalize the data into a spreadsheet, highlight price changes above a given threshold and post the final table to her team’s chat channel. Minutes later, she reviews the output without manually collecting screenshots or copying and pasting figures. In another case, an HR professional asks for a compliance report of required safety training modules; Runner H retrieves completion status from learning management systems, flags missing entries and generates reminder notifications for the relevant employees.

To support reliable browser interactions, H Company released Holo-1, a family of vision-language models in 3-billion and 7-billion-parameter variants. When paired with Surfer H, Holo-1 achieves a 92.2% success rate on the WebVoyager benchmark, which measures an agent’s ability to identify and interact with web-interface elements. The model remains compact enough to run on a single GPU, keeping inference costs low. All model weights and the 1,639-scenario WebClick dataset are now available on Hugging Face, providing developers and research groups with resources for fine-tuning or training agents that can read, interpret and click through complex digital interfaces.

When development teams push new features at faster cadences, manual test plans often lag behind. Tester H, currently in private beta, converts user stories written in everyday language into fully executable test suites. The agent parses each story, identifies relevant UI elements, constructs test scripts, runs them in a headless browser and captures screenshots, logs and performance metrics for any failures. Integration with continuous integration pipelines and reporting dashboards delivers rapid feedback. Early adopters report shorter regression cycles and greater confidence in deployment stability.

Beyond these initial tools, H Company outlines a roadmap for an expanding developer ecosystem. In the coming months, the firm plans to release a software development kit that lets third-party teams build and register custom sub-agents, extending the core platform to industry-specific tasks. A curated marketplace will offer prebuilt workflows—such as finance reconciliation, legal contract analysis and supply-chain monitoring—ready for immediate use or easy adaptation. Built-in analytics dashboards will surface performance metrics like task success rates, average execution time and resource consumption, giving organizations insight into workflow efficiency and areas that may require retraining or optimization.

To foster community engagement, H Company has opened developer forums and research collaborations around its open-sourced vision model. Early adopters, academic partners and independent contributors can log issues, propose model refinements or share custom agent templates in a public repository. Documentation includes code samples, best-practice guides and use-case libraries. A series of webinars and written tutorials aims to help teams onboard new features quickly while growing an active ecosystem of agent developers and power users.

On the product roadmap, H Company plans to introduce a conversational interface that lets users interact with Runner H through voice assistants and chatbots. A forthcoming mobile application will bring agent-driven automation to smartphones and tablets, bringing on-the-go execution of approved workflows. Real-time collaboration features will allow multiple team members to observe or annotate active workflows, propose updates and reroute tasks without interrupting the overall sequence. Enterprise customers will receive compliance modules that generate audit trails of data access, agent decisions and execution timestamps, helping to meet regulatory standards across industries.

H Company views these launches as foundational for moving agentic AI from isolated demonstrations into standard toolchains. By combining a flexible orchestrator with an open-licensed vision model and a natural-language–driven testing engine, the firm aims to reduce manual overhead and accelerate digital workflows. Observers in the tech community will be watching adoption rates and real-world performance as enterprises evaluate whether this integrated approach can deliver sustainable productivity gains at enterprise scale.