ToolLIFT: Lifting Tool-Specific Trajectories into Function-Level Graphs for Generalizable Tool Planning

2026-08-04

AI Summary

Existing approaches rely on tool-level graph representations of historical trajectories, which struggle to generalize to new tool sets and thereby limit the planning capabilities of large language models. To address this, this work proposes a Functional-level Workflow Graph (FWG) that abstracts tool-specific behaviors into functional-level workflows through trajectory uplifting, effectively decoupling workflow planning from tool selection. The framework incorporates a source-gating mechanism and skill-specific rewards, combined with reinforcement learning, to ensure reliable and traceable data flows. Evaluated on two in-distribution and three out-of-distribution benchmarks, the method significantly outperforms current state-of-the-art approaches and demonstrates strong cross-domain generalization to unseen tool sets.

Abstract

Historical tool-use trajectories provide valuable experience for large language model (LLM) agents to plan and coordinate tool usage. Existing approaches directly construct tool-level graphs from these trajectories, but the resulting graphs remain tied to specific tools and are hard to generalize across tool sets. To tackle this challenge, we find that despite differences in the tools involved, analogous tasks often share a common function-level workflow structure, which serves as a potentially more transferable abstraction for tool planning. Based on this insight, we propose ToolLIFT, a framework that lifts tool-specific trajectories into a function-level workflow graph (FWG) for generalizable tool planning. Specifically, we first propose a trajectory-lifting mechanism that encodes workflow structures in the FWG and shares collaboration experience across tools. Then, building on the global structure of the FWG, we introduce decoupled workflow planning and tool selection to align individual tool choices with the overall workflow. Lastly, to ensure reliable tool dataflow, we adopt Reinforcement Learning (RL) and propose source-gated and skill-specific rewards to maintain source-traceable information flow across tool calls. Experiments on two in-distribution (ID) and three out-of-distribution (OOD) benchmarks show that ToolLIFT consistently outperforms state-of-the-art baselines, demonstrating strong generalization to unseen tool sets.

Problem

• tool planning
• generalization
• workflow abstraction
• tool-use trajectories
• function-level graph

Innovation

• function-level workflow graph
• trajectory lifting
• generalizable tool planning
• decoupled planning and selection
• reinforcement learning with source-gated rewards