FitText: Evolving Agent Tool Ecologies via Memetic Retrieval

Kyle Zheng, Han Zhang, Renliang Sun, Chenchen Ye, Wei Wang · 2026

Read FitText: Evolving Agent Tool Ecologies via Memetic Retrieval by Kyle Zheng, Han Zhang, Renliang Sun, Chenchen Ye, Wei Wang (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-2605-02411.txt.

Description

Kyle Zheng and four co-authors introduce FitText, a method that helps software agents find the right tool among tens of thousands of online services. A fixed search from the first request can miss the needed tool, and no plan can fix a tool that was never found. FitText treats search as a task with a budget, done while the agent works.

The method writes, tests, and refines short fake tool descriptions used as search probes. Its strongest form adds selection, local tuning, and memory of past probes, in the style of evolution. On a bench with more than sixteen thousand services, it reached a pooled pass rate well above fixed search. Gains were largest on vague tasks that need several tools.

Catalog record

CREATORKyle Zheng, Han Zhang, Renliang Sun, Chenchen Ye, Wei Wang
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE75 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.02411

Browse the full library · About the Tower of Babel