ToolOmni: Enabling Open-World Tool Use via Agentic learning with Proactive Retrieval and Grounded Execution
Shouzheng Huang, Meishan Zhang, Baotian Hu, Min Zhang Harbin Institute of Technology (Shenzhen) [email protected], [email protected],{hubaotian, zhangmin2021}@hit.edu.cn https://github.com/Huangsz2021/ToolOmni · 2026
Read ToolOmni: Enabling Open-World Tool Use via Agentic learning with Proactive Retrieval and Grounded Execution by Shouzheng Huang, Meishan Zhang, Baotian Hu, Min Zhang Harbin Institute of Technology (Shenzhen) [email protected], [email protected],{hubaotian, zhangmin2021}@hit.edu.cn https://github.com/Huangsz2021/ToolOmni (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-13787.txt.
Description
Huang and three colleagues enable language models to use tools in open-world use with massive tool repositories. Static retrieval misaligns intent with tool meaning. Memorizing tools fails on unseen ones. Their ToolOmni framework combines proactive retrieval with grounded use inside a reasoning loop. In plain terms, the model must pick the right tool and use it well. A cold-start dataset teaches skills by supervised tuning. Then a decoupled GRPO tunes retrieval and execution together. ToolOmni leads baselines in both, lifting execution success by 10.8 percent.
Catalog record
| CREATOR | Shouzheng Huang, Meishan Zhang, Baotian Hu, Min Zhang Harbin Institute of Technology (Shenzhen) [email protected], [email protected],{hubaotian, zhangmin2021}@hit.edu.cn https://github.com/Huangsz2021/ToolOmni |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 63210 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.13787