From intent to data: An LLM-agentic framework for 6G automated experimentation

The increasing complexity of 6G experimental infrastructures presents significant challenges for researchers seeking to discover and generate datasets on heterogeneous testbeds and digital twins. Traditional approaches require a deep familiarity with platform-specific APIs, configuration schemas, and measurement pipelines, creating substantial barriers to adoption. This paper introduces a novel LLM-based agentic framework that enables intent-driven data search and automated experiment generation across distributed 6G experimental environments. The proposed system employs a cooperative ensemble of specialized agents, including: Chatbot, Search, Experiment, Translation, and Executor agents, that progressively translate high-level natural language user intents into concrete, executable experiment workflows. The framework integrates semantic scenario matching using sentence transformer embeddings, Model Context Protocol (MCP) for standardized tool interoperability, and a testbedadaptive translation layer that decouples intent interpretation from platform-specific execution mechanisms. The paper further defines a comprehensive evaluation methodology covering intent understanding, dataset retrieval, semantic scenario matching, end-to-end pipeline reliability, and guardrail effectiveness, providing a rigorous basis for future experimental validation. The proposed architecture addresses critical challenges in experiment portability, reproducibility, and accessibility, establishing a foundation for intent-driven experimentation in next-generation network research.