So we shifted focus. As an alternative of forcing generative AI into fragmented workflows, we got down to design a platform that felt native to our current setting. That led to LMOS, the Language Mannequin Working System, a sovereign PaaS for constructing and scaling AI brokers throughout Deutsche Telekom. LMOS gives a Heroku-like expertise for brokers, abstracting away life-cycle administration, deployment fashions, classifiers, observability, and scaling whereas supporting versioning, multitenancy, and enterprise-grade reliability.
On the core of LMOS is Arc, a Kotlin-based framework for outlining agent habits by means of a concise domain-specific language (DSL). Engineers might construct brokers utilizing the APIs and libraries they already knew. No must introduce totally new stacks or rewire improvement workflows. On the identical time, Arc was constructed to combine cleanly with current knowledge science instruments, making it straightforward to plug in customized elements for analysis, fine-tuning, or experimentation the place wanted.
Arc additionally launched ADL (Agent Definition Language), which permits enterprise groups to outline agent logic and workflows immediately, decreasing the necessity for engineering involvement in each iteration and enabling sooner collaboration throughout roles. Collectively, LMOS Arc, and ADL helped bridge the hole between enterprise and engineering, whereas integrating cleanly with open requirements and knowledge science instruments, accelerating how brokers have been constructed, iterated, and deployed throughout the group.
