Declarative Data Services: Structured Agentic Discovery for Data Systems
A recent study published on arXiv (2605.20690) introduces Declarative Data Services (DDS), a framework designed for the structured agentic exploration of data-system compositions based on user intent. This architecture tackles the complexities of integrating diverse multi-system data backends, where the search environment is varied and the knowledge for composition is inconsistently represented in pretraining. DDS employs four distinct contracts across different layers—intent, operator DAG, per-system capabilities, and runtime attribution—to break down the overall search into manageable sub-searches. Each sub-agent investigates its specific typed area, while the framework offers organized direction. The authors highlight that unbounded agentic discovery, which relies on iterative failure-log feedback, often struggles to consistently yield a functional stack, even with iteration and clear composition knowledge. DDS seeks to enhance the reliability and efficiency of discovering innovative data system compositions.
Key facts
- arXiv paper 2605.20690 proposes Declarative Data Services (DDS).
- DDS is an architecture for structured agentic discovery of data-system compositions.
- It addresses heterogeneous search spaces and uneven composition knowledge in pretraining.
- Unbounded agentic discovery fails to converge consistently on a working stack.
- DDS uses four typed contracts: intent, operator DAG, per-system skills, runtime attribution.
- Sub-agents search each typed space with framework guidance.
- The paper is categorized under cs.DB and cs.AI.
- The approach decomposes global search into bounded sub-searches.
Entities
Institutions
- arXiv