The idea behind proxy AI is that many small mission-centric agents can collaborate to get the actual work done. However, this particular idea is more like a hope than the product. Fortunately, the Paris-based H company wants to change that, announcing bringing our vision for proxy AI to life, starting with putting its flagship runners on the starting line.
H’s runner H is now in a public beta, inviting anyone to fire a prompt and watch a series of sub-agents to fill spreadsheets, scrape websites, Ping Slack, and even install invoices when scanning results.
The Runner H distribution appeared along with two other announcements: the Open Vision Model, which directed Runner H’s browser cousin Surfer H and launched a private BETA test platform called HET Tester H. Together, these three announcements draw a practical roadmap for bringing multi-faceted systems into daily engineering.
What is Runner H?
According to H, Runner H is the most advanced AI agent that allows anyone to automate complex, tedious multi-step tasks without repeated and manual input.
In short, Runner H is the coordinator for your digital workloads:
You provide an advanced goal and the internal orchestration system breaks. The system intelligently assigns tasks to dedicated sub-agents, including their browsing agents, surfer H, and other connected applications. This allows runners to:
- Extract real-time data, compile it into a spreadsheet, and share it with your teammates via Slack.
- Search online for related work open and then continue the application process.
- Update your Customer Relationship Management (CRM) system with new lead information and a custom follow-up email draft for that lead.
Runner H’s design answers two frustrations that developers mentioned most using large language models (LLMs): fragmented context and weak execution.
Instead of returning the suggested paragraph, Runner H assigns the task to a mini proxy that can plan, invoke the API, click on the UI and track what happened last time. Company H says this methodmemory + Orchestration + implement. “In practice, this means that tasks used to bounce between shell scripts, Zapier and human inspectors can live in a chat thread.
Runner H integrates memory, task orchestration, execution functions, and deep connections to a single interface with other software.
Here are the key features:
- Player arrangement: Users can use a prompt to start the complete workflow. Runner H then coordinates the different AI agents in a synchronous manner to plan and execute the necessary steps.
- Application Integration: It connects to popular tools such as Slack, Intion, Google Sheets, and various APIs, allowing it to run directly in your existing software settings for immediate automation.
- Knowledge upload: Users can provide PDFs, documents and data files. Runner H processes this information, converting it into accessible contexts to inform its actions and produce more accurate results.
- Automatic payment (coming soon): H Company plans to allow Runner H agents to purchase or subscribe to software services from platforms such as Amazon and Shopify without direct human intervention.
According to H Company, Runner H aims to redefine interactions with AI by creating a more intuitive and efficient way of digital tasks.
How to use Runner H:
Step 1: Visit the Runner H webpage on the H company website and click “Try”.
Step 2: Register to start. Enter your prompt and submit.
- Quickly: Please visit Marktechpost.com and summarize 3 latest AI research news articles.
Step 3: Runner H will take several minutes to complete the task, depending on the complexity of the task. You can see and track everything from beginning to end.
Holo-1: Open weight of visual navigation
An automated browser means viewing the browser. To make this reliable and cheap, H Company released Holo-1, a family of 3 billion and 7 billion parameter action visual language models.
Pairing H with Surfer H, Holo-1 gets 92.2% UI localization in the WebVoyager benchmark while keeping it small enough to run on a single GPU. The weights and 1,639-Scenario’s WebClick dataset are both live in the hugging face, providing fresh material for academics and startups for training agents that can be read and clicked.
Tester H: Close quality inspection gap
If the AI launches code in seconds, quality assurance cannot rely on weekly manual testing plans. Tester H (now the private beta of the enterprise) turns pure English user stories into executable end-to-end testing. ‘
The agent will cross the interface like a real customer and then start clicking the button, opening mode, checking text and layout, and reporting any deviations.
Early tests show that they have cut off the regression cycle while increasing confidence in each push. Requests to join Beta are open on H Company’s website.
in conclusion
The recent announcement from H Company, especially the announcement of launching Runner H into the public beta, is the first open playground that points to a future where AI agents play a more active role in personal and professional productivity. Adding open source vision models to Holo-1 (such as HOLO-1) and test reagent H provides a valuable resource for the wider AI community to target DevOps pain points directly.
Company H dares to bet that 2025 will be the agents no longer a demonstration and begin to become part of the toolchain. Currently, the introduction of more complex proxy AI capabilities is an interesting development to watch in the ever-evolving technical stories.
Nishant, product growth manager at Marktechpost, is interested in learning about artificial intelligence (AI), what it can do and what it develops. His passion for trying new things and giving it creative twists helped him interact with Tech. He is assisting the company in promoting growth and market recognition.
