AgentSM : Semantic Memory for Agentic Text-to-SQL

Asim Biswal △∘ , Chuan Lei ∗⋄ , Xiao Qin ∗□ , Aodong Li ∘ , Balakrishnan Narayanaswamy ∘ , Tim Kraska ∘ · 2026

Read AgentSM : Semantic Memory for Agentic Text-to-SQL by Asim Biswal △∘ , Chuan Lei ∗⋄ , Xiao Qin ∗□ , Aodong Li ∘ , Balakrishnan Narayanaswamy ∘ , Tim Kraska ∘ (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-15709.txt.

Description

Asim Biswal and five co-authors give language models a memory for database work in a 2026 arXiv paper. Their system is AgentSM, a semantic memory for text-to-SQL agents. Text-to-SQL turns plain questions into database queries. Public benchmarks like BIRD and Spider show strong recent gains. Real enterprise settings are harder. Schemas are large and complex, dialects vary, and reasoning runs in costly steps.

A semantic memory can store what the agent learns across queries. Past schemas, joins, and fixes become reusable knowledge. That should cut repeated reasoning and steady the output. The work was done at Amazon Web Services, per a note in the record. For firms with sprawling data estates, an agent that remembers its own experience is the practical draw.

Catalog record

CREATORAsim Biswal △∘ , Chuan Lei ∗⋄ , Xiao Qin ∗□ , Aodong Li ∘ , Balakrishnan Narayanaswamy ∘ , Tim Kraska ∘
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE42 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-15709

Browse the full library · About the Tower of Babel