A Language for Describing Agentic LLM Contexts (arxiv.org)

A recent development in the AI community is the introduction of the Agentic Context Description Language (ACDL), designed to provide a standardized way to specify the structure and evolution of contexts used in large language model (LLM) systems. As LLMs become integral to complex agentic systems that involve multiple interactions, effectively communicating the composition of context—how it changes across interactions—is crucial for system performance. ACDL addresses the current lack of precision in conveying context constructs, which has previously relied on informal methods like prose or diagrams. ACDL allows for a more structured and visual representation of context aspects, including role message sequences, dynamic content, and iterative logic, facilitating clearer communication and understanding among researchers and developers. By documenting existing systems and encouraging its adoption in academic papers and daily interactions, ACDL aims to enhance the technical dialogue within the AI/ML community. With available tooling and documentation, it stresses the importance of consistency and clarity in LLM context engineering, paving the way for improved design and collaboration in the development of LLM agents.