Newsroom

Navigating interfaces. Reading and interpreting documents. Clicking on the right components. And repeat. Currently, these tasks are done every day, sometimes all day, by humans.

Until now.

Today we're introducing the Studio, our platform for developers, and eventually anyone, to build reliable automations without much effort. It's extensible, and over time it will handle more kinds of automation and reach beyond the web.

Our main agent, Runner H, will be in private beta on the platform. You describe what you want in plain language and it builds the web automation pipeline for you, taking over the tedious multi-step work that web testing and process automation usually involve.

As a web agent, Runner H delivers state-of-the-art performance, outperforming Anthropic Computer Use on the public benchmark WebVoyager (learn more). Our in-house foundation models, smaller, specialized, and cheaper, potentially by orders of magnitude, can outperform large generalist models when powering agents.

We're excited for Runner H to help kick off a wave of agents that can reliably handle complex tasks. If you want to run with us, sign up for the private beta waitlist.

The Studio and Runner H 0.1

Web developers lose hours maintaining brittle selectors and fixing automations that break every time a web interface changes.

Runner H, our web agent in the Studio, handles this: it follows natural-language instructions, adapts to UI changes on its own, and repairs itself when something breaks.

While our agent handles the complexities of writing and maintaining selectors behind the scenes, developers can focus on the semantics of the workflows and production, freeing up time for higher value development work.

The private beta, coming soon, includes:

• the API to call off-the-shelf and managed agents running in the cloud
• the Studio to create automations, review and edit past and live runs

In the Studio, you can build reliable automations for complex workflows like end-to-end e-commerce scenarios and testing (from product discovery to order confirmation) and financial services onboarding (pre-filling multi-step verification processes, document uploads, and compliance checks).

Big strides toward a big vision: plan, see, run & repeat

Runner H can hit production-grade automation more reliably than older approaches like screen scraping, and it does it at scale. This comes out of what our engineering and research teams have built since the company started:

• We have trained the best Vision Language Model (VLM) on the market for predicting the coordinates of the mouse click when given a screenshot and a natural language instruction such as "Click the add to cart button," as shown by our performance in Screenspot, a benchmark specialized in UI for Action Models.
• We have designed the best web agent on the market for open-ended tasks, as shown by our performance on WebVoyager.

To understand the foundation of Runner H, learn more about our VLMs and LLMs here.

The road ahead

Ultimately, we envision a future where you can interact with Runner H as naturally as you would with a colleague.

Going forward, we will advance on several fronts:

• Improving accuracy and cost efficiency through complex large-scale techniques including reinforcement learning and distillation
• Adding debugging and teaching capabilities to the Studio so developers can train Runner H 0.1 to perform well on their specific tasks
• Fostering a developer community through technical content, support, and events
• Upholding enterprise-grade security standards and ensuring Runner H 0.1 operates safely and reliably (more to come on our research)

This is just the first step toward making agents widely usable. Less time spent on tedious tasks means more time for the work that actually matters.

Join us in shaping where web automation goes next.

Sign up for the private beta waitlist.