AgentSM: Semantic Memory for Agentic Text-to-SQL

TL;DR

AgentSM introduces an interpretable semantic memory framework for Text-to-SQL tasks, enhancing scalability, efficiency, and accuracy in complex enterprise scenarios by reusing reasoning paths and guiding future queries.

Contribution

This paper presents AgentSM, a novel agentic framework that leverages structured semantic memory to improve reasoning efficiency and accuracy in Text-to-SQL systems.

Findings

Achieves 25% reduction in token usage and 35% shorter trajectories compared to state-of-the-art.

Reaches 44.8% accuracy on Spider 2.0 Lite benchmark, setting a new state-of-the-art.

Effectively scales to larger schemas and more complex questions.

Abstract

Recent advances in LLM-based Text-to-SQL have achieved remarkable gains on public benchmarks such as BIRD and Spider. Yet, these systems struggle to scale in realistic enterprise settings with large, complex schemas, diverse SQL dialects, and expensive multi-step reasoning. Emerging agentic approaches show potential for adaptive reasoning but often suffer from inefficiency and instability-repeating interactions with databases, producing inconsistent outputs, and occasionally failing to generate valid answers. To address these challenges, we introduce Agent Semantic Memory (AgentSM), an agentic framework for Text-to-SQL that builds and leverages interpretable semantic memory. Instead of relying on raw scratchpads or vector retrieval, AgentSM captures prior execution traces-or synthesizes curated ones-as structured programs that directly guide future reasoning. This design enables systematic reuse of reasoning paths, which allows agents to scale to larger schemas, more complex questions, and longer trajectories efficiently and reliably. Compared to state-of-the-art systems, AgentSM achieves higher efficiency by reducing average token usage and trajectory length by 25% and 35%, respectively, on the Spider 2.0 benchmark. It also improves execution accuracy, reaching a state-of-the-art accuracy of 44.8% on the Spider 2.0 Lite benchmark.