FitText: New Framework Improves AI Agent Tool Selection Through Dynamic Retrieval
Researchers introduced FitText, a training-free framework that improves how AI agents find and use the right tools from large API ecosystems by dynamically refining tool descriptions during task execution. The system treats tool retrieval as an evolving process where agents generate and iteratively refine natural-language descriptions of needed tools rather than relying on static initial queries. The approach achieved significant performance gains, including a 26.7-point improvement over traditional retrieval methods on a benchmark with over 16,000 APIs.
FitText addresses a fundamental challenge in scaling AI agents: as API ecosystems grow to tens of thousands of endpoints, agents struggle to find the right tools because there is a semantic gap between how users describe tasks and how tools are documented. The framework makes tool retrieval dynamic by embedding it directly into the agent's reasoning loop, allowing the agent's understanding of needed tools to evolve during execution. Rather than requiring training, FitText uses a memetic retrieval approach where agents generate pseudo-tool descriptions as revisable hypotheses, refine them iteratively using retrieval feedback, and explore alternatives through stochastic generation. A tool memory component adds evolutionary selection pressure to avoid redundant searches. Testing on ToolRet and StableToolBench (which contains 16,464 APIs) showed improvements of 2.7 to 10.6 points in NDCG@5 metrics across different base models, with GPT-4-mini achieving an 84.3% pass rate—a 26.7-point absolute gain over static retrieval.
What's missing
The paper does not discuss computational overhead or latency implications of the iterative refinement process compared to static retrieval, which would be relevant for real-world deployment. Additionally, the generalization of FitText to domains beyond the tested benchmarks and its performance with different types of tool ecosystems (e.g., non-API tools) remains unclear.
What different sources said
- arXiv cs.AICenter
FitText: Evolving Agent Tool Ecologies via Memetic Retrieval
Related
Genetic Drift, Not Selection, Drives Rapid Feather Color Evolution in Island Bird Radiation
A new study of an island bird radiation found that rapid evolution of feather coloration is driven primarily by genetic drift in small populations rather than sexual or ecological selection. The research integrated whole-genome data with detailed plumage measurements across complete species sampling to test whether signaling trait evolution correlates with speciation rates. The findings suggest that neutral demographic processes play a central role in generating phenotypic diversity during island radiations, challenging assumptions about the mechanisms driving rapid evolution.
New AI Model Improves Prediction of Therapeutic Peptide Function from Protein Sequences
Researchers developed a lightweight CNN classifier that predicts whether peptide sequences have therapeutic properties, trained on a database of 54,655 peptides across 48 functional categories. The model uses a novel negative sampling strategy to reduce false positive rates from over 60% in previous approaches to 2.1%. This advancement could accelerate drug discovery by enabling faster computational screening of peptide candidates before expensive experimental testing.
Study Shows Different Metabolic Stress Models Produce Distinct Effects on Human Neuronal Networks
Researchers tested three common in vitro metabolic stress models on human-derived neuronal networks and found each produced different patterns of neuronal activity and cell damage. The models tested were hypoxia alone, oxygen-glucose deprivation (OGD), and hypoxia combined with glutamate exposure. The findings suggest that choice of experimental model significantly affects results and that combining electrophysiological and structural analyses is important for accurately assessing metabolic stress in stroke research.