Tracer improves tool choices while cutting input tokens for AI agents

TRACER: Trajectory-Aware Tool Retrieval for LLM Agents

Computation and LanguageArtificial IntelligenceInformation Retrieval

Summary

Modern AI agents use many tools but listing all possible tools can be slow and costly as the list grows. The authors created TRACER, which remembers sequences of past tool uses and smartly suggests the next tools, showing only a small part of the full list. Their method makes AI agents like GPT-5.4 pick tools more accurately while using far fewer input tokens. They tested TRACER in different environments and with several AI models, finding it usually improves tool choices compared to showing all tools. They also studied when the system fails to help improve future versions.

What this means in practice

Authors

Tanish Kher, Sarvesh DK, Abedelkadir Asi, Sparsha Chennarayapatna Sundaresh, Dezhi Hong

Abstract

Modern agents select tools from registries, but exposing entire registries incurs high token costs and can impair selection as they grow. We present TRACER, a retrieval front-end that indexes observed tool-call sequences in a trie and ranks continuations using a learned blend of empirical and contextual priors. On a mixed retail-and-telecom $τ^2$ environment, TRACER improves GPT-5.4's tool-selection accuracy by 4.7 points while reducing input tokens per step by 80%, exposing just 5 of 128 tools. In autonomous full-turn execution, it matches GPT-5.4's full-registry task success with 61% fewer input tokens per step. Across six environments and four models, TRACER improves tool-selection accuracy over full-registry selection in 21 of 24 model-environment pairs, with average gains of 6 to 10 points; the three exceptions trail by at most 2.2. We analyze failures in both regimes to identify directions for future work.