SAG: SQL-Retrieval Augmented Generation with Query-Time Dynamic Hyperedges
For practitioners needing efficient multi-hop retrieval over structured data, SAG offers a scalable, maintenance-free alternative to knowledge-graph-based RAG.
SAG introduces a SQL-based retrieval architecture that dynamically constructs local hyperedges at query time, avoiding global graph maintenance. It achieves state-of-the-art results on 8 out of 9 Recall@K metrics across three multi-hop benchmarks, reaching 80.0% Recall@5 on MuSiQue, and scales to hundreds of millions of items with sub-second latency.
Retrieval-Augmented Generation (RAG) offers an effective approach for large language models to access external knowledge. However, existing methods rely on dense similarity retrieval and face inherent limitations in handling structured constraints and multi-hop reasoning. Incorporating knowledge graphs partially alleviates these issues, but at the cost of semantic fragmentation, high maintenance overhead, and difficult incremental updates. This paper introduces SAG (SQLRetrieval Augmented Generation), a structured architecture for retrieval and agent systems. Instead of pre-building a global static graph, SAG converts each chunk into one semantically complete event and a set of indexing entities, then uses SQL join queries to dynamically link events that share entities into local hyperedges,constructing, at query time, a dynamically instantiated local index structure. This design avoids the need for global graph rebuilding and ongoing maintenance; the system naturally supports incremental writes, concurrent processing, and continuous scaling through its reliance on standard database infrastructure. Across HotpotQA, 2WikiMultiHop, and MuSiQue, three standard multi-hop benchmarks,SAG achieves the best results on 8 out of 9 Recall@K metrics, reaching 80.0% Recall@5 on MuSiQue, the benchmark with the highest multi-hop reasoning demands.SAG has also been deployed at a production scale of hundreds of millions of data items, with online retrieval latency kept within seconds. Project site and code are available at https://github.com/Zleap-AI/SAG-Benchmark.