Think-Augmented Function Calling: Improving LLM Parameter Accuracy Through Embedded Reasoning

TL;DR

Think-Augmented Function Calling (TAFC) introduces explicit, parameter-level reasoning within function calling for LLMs by augmenting function signatures with a universal think parameter while preserving execution structure. The framework adds granularity through reasoning-enhanced parameter tuples and a complexity-based trigger to generate justification only when needed, coupled with dynamic description tuning and reasoning-guided tool optimization to align model traces with human expectations. TAFC operates without architectural changes to LLMs, enabling seamless API compatibility and traceable reasoning via a centralized repository of interactions for continuous improvement. Evaluations on ToolBench across proprietary and open-source models demonstrate consistent improvements in parameter accuracy and reasoning coherence, particularly for complex multi-parameter functions, while providing enhanced interpretability for debugging autonomous agents.

Abstract

Large language models (LLMs) have demonstrated remarkable capabilities in function calling for autonomous agents, yet current mechanisms lack explicit reasoning transparency during parameter generation, particularly for complex functions with interdependent parameters. While existing approaches like chain-of-thought prompting operate at the agent level, they fail to provide fine-grained reasoning guidance for individual function parameters. To address these limitations, we propose Think-Augmented Function Calling (TAFC), a novel framework that enhances function calling accuracy through explicit reasoning at both function and parameter levels. Our method introduces a universal "think" parameter augmentation that enables models to articulate their decision-making process, with dynamic optimization for parameter descriptions to improve reasoning quality. For complex parameters, TAFC automatically triggers granular reasoning based on complexity scoring, ensuring appropriate justification for critical decisions. Additionally, we propose reasoning-guided optimization to align generated reasoning with human expectations. TAFC requires no architectural modifications to existing LLMs while maintaining full API compatibility. Evaluation on ToolBench across proprietary and open-source models demonstrates significant improvements in parameter generation accuracy and reasoning coherence for multi-parameter functions, while providing enhanced interpretability for debugging AI agent behaviors.