SynthTools: A Framework for Scaling Synthetic Tools for Agent Development

ICLR 2026 Conference Submission22611 Authors

Keywords: LLM Agents, Tool Generation, Scalable Environments

Abstract: AI agents increasingly rely on external tools to solve complex, long-horizon tasks. Effective development of such agents requires large-scale training in environments where they can safely practice using diverse tools, adapt strategies, and iteratively improve. However, real-world APIs are limited in availability, domain coverage, and stability, often requiring access keys and imposing rate limits, rendering them impractical for scalable training. To address these challenges, we introduce SynthTools, a flexible and scalable framework for generating synthetic tool ecosystems. Our framework consists of three core components: Tool Generation for automatic and scalable creation of diverse tools across domains, Tool Simulation to emulate realistic tool behaviors, and Tool Audit to ensure consistency and reliability. Using SynthTools, we generate large corpora of synthetic tools and tasks, enabling controllable, stable, and domain-agnostic training environments for LLM agents. By decoupling training from real-world API constraints, SynthTools provides stable interfaces, supports multi-domain experimentation, and thereby accelerates the development of robust, general-purpose LLM agents.

Primary Area: foundation or frontier models, including LLMs

Submission Number: 22611