In short

A database that treats LLM prompts as a type of relational data, allowing the system to rewrite, optimize and manage them natively.

Martins, D. M. L. & Vossen, G. (2026)

Summary

Language Models (LLMs) are increasingly used with relational databases, but prompts are usually isolated outside the bank (in code), invisible for the optimization of query and metadata. The “PromptDB” introduces the type of logical data PROMPT, allowing the database optimizer to rewrite the prompts to improve the cost-benefit and accuracy of the use of LLMs.

Discussion

Open track for the next club meeting.

View on arXiv

Automatic translation notice

This page was automatically translated from Portuguese using the LibreTranslate-based automated translator implemented in tools/translate_quartz.py (it preserves wikilinks, embeds and proper names via positional splitting). Machine translation may contain inaccuracies — the original Portuguese version is the authoritative source.



© 2026 Pedro Henrique Rocha de Andrade · Built with Quartz