Orby AI

Orby AI is a generative process automation platform built on a proprietary Large Action Model, founded in Mountain View by chief executive Bella Liu, who led AI product at UiPath through its public offering, and chief technology officer Will Lu, formerly head of cloud AI engineering at Google. It raised thirty five million dollars from investors including New Enterprise Associates, Pear, Wing, and WndrCo, and deployed to Fortune 50 and Fortune 500 companies for mission critical work like finance auditing and legal contract review. In early 2026 Orby was acquired by Uniphore, folding its research and technology into Uniphore's Business AI Cloud.

Orby's core is its Large Action Model, called ActIO, which the company describes as the industry's first large action model foundation for the enterprise. Unlike language models that interpret text, a large action model observes actions. ActIO watches a user work, analyzing clicks, forms, buttons, keystrokes, screenshots, and page structure, then generates the code to automate that workflow, adapting as circumstances change. This is paired with a visual grounding model, developed with Ohio State University and open sourced, that connects what the agent sees on screen to what it understands, giving it genuine command of graphical interfaces.

The company calls its category generative process automation, positioning it as a no rules alternative to rule based robotic process automation. Where legacy automation breaks when a screen changes, Orby's model reasons about intent and handles exceptions, breaking complex processes into smaller tasks and executing them with minimal supervision. It works across any application by observing the interface rather than requiring deep custom connectors, and it orchestrates specialized agents for sub tasks like data analysis. A Fortune 500 finance team used it to move from auditing ten percent of expense receipts to one hundred percent, saving four million dollars a year.

Orby uses a Guided Autonomy framework, where agents first work alongside human reviewers and require approval, with oversight decreasing as the model learns from feedback. Its greatest strength is genuine computer use, driven by that action model and visual grounding, plus autonomous workflow generation. As a focused automation engine rather than a broad enterprise platform, several capabilities were not documented, including named security certifications, on premises deployment, and a public extensibility surface. Following the Uniphore acquisition, its technology now advances within a larger business AI platform.