Deutsche Telekom's Open-Source LMOS Platform Powers AI Agents at Scale
Explore Deutsche Telekom's LMOS platform and its Kotlin-based Arc framework for building, deploying, and managing cloud-native AI agent systems.
Deutsche Telekom has developed an open-source platform, LMOS (Language Model Operating System), to deploy AI-powered assistants across its multi-country ecosystem. The initiative, led by the AI Competence Center (AICC), aims to enhance customer service efficiency and reduce resolution times. The platform supports millions of interactions, including the customer-facing assistant Frag Magenta OneBOT, which handles sales and service queries.
Sovereign AI Infrastructure
In early 2023, the team recognized the limitations of relying on fragmented tools and third-party platforms for generative AI. Instead, they built LMOS as a sovereign PaaS grounded in open standards. The platform abstracts complexities like deployment, monitoring, and scaling, allowing engineers to focus on agent development using familiar JVM-based systems and APIs.
Key Components:
- Arc: A Kotlin-based framework with a domain-specific language (DSL) for defining agent behavior.
- ADL (Agent Definition Language): Enables business teams to define workflows without heavy engineering involvement.
- Qdrant: A Rust-based vector database for efficient, multi-tenant retrieval.
- Wurzel: An open-source Python ETL framework for structuring RAG (Retrieval-Augmented Generation) pipelines.
Semantic Routing and Retrieval
LMOS integrates vector search to provide agents with contextual knowledge from internal documentation and policies. Semantic routing classifies and directs queries (e.g., billing, sales) using embeddings, reducing reliance on LLMs for basic tasks. This approach improves precision and interpretability.
Results and Future Outlook
- Reduced development time: New agents can be built in a day or less.
- Lower handover rates: Human intervention is now around 30%, with further improvements expected.
- Open-source contribution: LMOS was donated to the Eclipse Foundation to foster community collaboration.
Deutsche Telekom’s focus on developer experience and interoperable architecture has positioned LMOS as a scalable solution for enterprise AI. The platform’s success underscores the potential of agentic AI in real-world applications.
For more on generative AI, visit InfoWorld’s Generative AI Insights.
—Arun Joseph, former engineering lead at Deutsche Telekom.
Related News
Data Scientists Embrace AI Agents to Automate Workflows in 2025
How data scientists are leveraging AI agents to streamline A/B testing and analysis, reducing manual effort and improving efficiency.
Agentic AI vs AI Agents Key Differences and Future Trends
Explore the distinctions between Agentic AI and AI agents, their advantages, disadvantages, and the future of multi-agent systems.
About the Author

Michael Rodriguez
AI Technology Journalist
Veteran technology journalist with 12 years of focus on AI industry reporting. Former AI section editor at TechCrunch, now freelance writer contributing in-depth AI industry analysis to renowned media outlets like Wired and The Verge. Has keen insights into AI startups and emerging technology trends.