Budget-Constrained Agentic Large Language Models: Intention-Based Planning for Costly Tool Use

TL;DR

This paper introduces INTENT, a planning framework for budget-constrained large language models that efficiently manages tool use costs and success rates in multi-step tasks by leveraging an intention-aware hierarchical model.

Contribution

The paper presents a novel inference-time planning method, INTENT, that effectively handles budget constraints and stochastic tool outcomes in large language model agents.

Findings

INTENT enforces strict budget feasibility.

It significantly improves task success rates.

The method remains robust under dynamic market conditions.

Abstract

We study budget-constrained tool-augmented agents, where a large language model must solve multi-step tasks by invoking external tools under a strict monetary budget. We formalize this setting as sequential decision making in context space with priced and stochastic tool executions, making direct planning intractable due to massive state-action spaces, high variance of outcomes and prohibitive exploration cost. To address these challenges, we propose INTENT, an inference-time planning framework that leverages an intention-aware hierarchical world model to anticipate future tool usage, risk-calibrated cost, and guide decisions online. Across cost-augmented StableToolBench, INTENT strictly enforces hard budget feasibility while substantially improving task success over baselines, and remains robust under dynamic market shifts such as tool price changes and varying budgets.