Natural AI Code Assistant
An AI-powered assistant that uses LLMs to generate, optimize, and modernize Natural code within the Adabas & Natural ecosystem.
The Natural AI Code Assistant brings large language models to the Adabas & Natural ecosystem, accelerating the modernization of enterprise mainframe applications. It is built with an API-first MVP that integrates cleanly into existing developer workflows and supports rapid prototyping against real customer feedback.
Under the hood, the assistant uses agentic workflows (built with LangChain and LangGraph) for dynamic, context-aware code generation, together with self-corrective strategies that iteratively improve the quality and reliability of generated Natural code. Semantic RAG pipelines ground the model in internal Natural documentation, improving contextual accuracy, while integration with MCP servers connects local and remote tooling to support hybrid deployment scenarios.
To achieve high-accuracy Natural code generation for enterprise workloads, several open-source LLMs (including Mistral, DeepSeek, and Qwen Coder) were fine-tuned for the language. The assistant was previewed to the Natural community through a series of Innovation Lab sessions at Software AG’s International User Group (IUG) conference, gathering hands-on feedback that continues to shape the product roadmap.