A new open-source framework is making it easier for developers to build, deploy, and manage an AI agent at scale.
Agentic AI promises systems that can autonomously reason, plan, and execute complex business tasks, moving automation beyond simple scripts. While hyperscalers like AWS, Google, and Microsoft – alongside enterprise platforms from IBM and SAP – are embedding agentic capabilities, familiar challenges remain: governance, scalability, and reliability.
How do you build, manage, and scale AI agents when their behaviour is defined by ambiguous natural language prompts? How do you version-control a prompt? How do you ensure an agent’s logic aligns with business rules when its definition is opaque, and how do you empower domain experts to participate in the design process?
According to Gartner, by 2028, 15 percent of daily business decisions will be made autonomously by agentic AI, with a third of all enterprise applications including such capabilities. This demands a new approach to agent design.
The Eclipse Foundation today offered a potential open-source answer to this AI problem, announcing the Agent Definition Language (ADL) and ARC Agent Framework as new components for its Eclipse LMOS (Language Models Operating System) project.
Eclipse LMOS is an open-source platform for orchestrating intelligent AI agents at enterprise scale. The ADL addition specifically addresses the weakness of prompt-based design by providing a model-neutral language and visual toolkit. The intent is to bridge the gap between business and engineering teams.
This approach aims to mature agent design into a more reliable engineering discipline. Using ADL, enterprises can co-define AI agent behaviour in a way that is consistent, maintainable, and versionable. This allows non-technical domain experts – including compliance officers, financial analysts, or logistics managers – to collaborate directly with engineers in shaping agent behaviour.
“With ADL, we wanted to make defining agent behaviour as intuitive as describing a business process, while retaining the rigor engineers expect,” said Arun Joseph, Eclipse LMOS project lead. “It eliminates the fragility of prompt-based design and gives enterprises a practical path to scale agentic AI using their existing teams and resources.”
This framework creates a clear alternative to the proprietary, black-box AI agent-building tools emerging from major tech vendors. The Eclipse Foundation is betting that enterprises will prefer a “sovereign, open platform” that avoids vendor lock-in and offers greater transparency.
“Agentic AI is redefining enterprise software, yet until now there has been no open source alternatives to proprietary offerings,” commented Mike Milinkovich, Executive Director of the Eclipse Foundation. “With Eclipse LMOS and ADL, we’re delivering a powerful, open platform that any organisation can use to build scalable, intelligent, and transparent agentic systems.”
Eclipse LMOS is already in production in one of Europe’s largest enterprise agentic AI deployments at Deutsche Telekom. The platform powers the ‘Frag Magenta OneBOT’ assistant and other customer-facing AI systems.
Deutsche Telekom’s deployment has successfully processed millions of service and sales interactions across several countries, demonstrating that the open-source model can meet enterprise-grade demands for scalability and reliability.
The enterprise appeal of LMOS is rooted in its architecture. LMOS is designed to integrate directly into existing enterprise IT environments. The platform itself is an orchestration layer built on the CNCF stack, including standards like Kubernetes and Istio. The AI agent framework (Eclipse LMOS ARC) is JVM-native with a Kotlin runtime; allowing organisations to leverage their deep existing investments in JVM-based applications, skills, and DevOps practices.
This combination is what the foundation sees as its unique value: it leverages existing engineering investments while simultaneously empowering business experts through ADL.
The emergence of platforms like Eclipse LMOS clarifies a central strategic choice. The decision is not just which large language model to use, but how to build the orchestration and governance layer that will manage the resulting agents.
Proprietary solutions from major vendors offer speed and deep integration within their ecosystems, but often at the cost of control and portability. The open-source path presented by LMOS and ADL offers a different value proposition: a modular, multi-tenant architecture on open standards that an enterprise can control and adapt itself.
The introduction of ADL provides a practical tool for solving the governance and reliability problem. By treating agent behaviour as a defined, versionable business process rather than an ambiguous prompt, the framework makes agentic AI auditable and scalable.
See also: EA partners with Stability AI for 3D asset creation
Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including the Cyber Security Expo, click here for more information.
Developer is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.