_version_ 1866911357439311872
author Chen, Jiawei
Shen, Xintian
Zheng, Lihao
Shao, Zhenwei
Cui, Handong
Du, Chaoqun
Gong, Li
Gu, Feng
Hao, Xuefeng
He, Wei
He, Jiabang
Hu, Yi
Huang, Bin
Li, Shanshan
Li, Qizhen
Luo, Jing
Liu, Zide
Liu, Xiaobo
Mao, Ning
Mu, Lifu
Pan, Xuhao
Qu, Zhiheng
Ren, Chang
Rao, Xudong
Sun, Haoyi
Wang, Qian
Wang, Shuai
Wang, Zhichao
Wang, Wei
Wen, Lian
Zhan, Jiqing
Yang, Hongfu
Yang, Sheng
Yang, Jiajun
Yu, Pengfei
Zhang, Hongyuan
Zhang, Bin
Zhou, Chunpeng
Zhou, Zheng
Zhou, Shucheng
Xie, Shuo
Zhu, Yun
Ma, Hao
Wei, Tao
Zhou, Pan
Chen, Wei
author_facet Chen, Jiawei
Shen, Xintian
Zheng, Lihao
Shao, Zhenwei
Cui, Handong
Du, Chaoqun
Gong, Li
Gu, Feng
Hao, Xuefeng
He, Wei
He, Jiabang
Hu, Yi
Huang, Bin
Li, Shanshan
Li, Qizhen
Luo, Jing
Liu, Zide
Liu, Xiaobo
Mao, Ning
Mu, Lifu
Pan, Xuhao
Qu, Zhiheng
Ren, Chang
Rao, Xudong
Sun, Haoyi
Wang, Qian
Wang, Shuai
Wang, Zhichao
Wang, Wei
Wen, Lian
Zhan, Jiqing
Yang, Hongfu
Yang, Sheng
Yang, Jiajun
Yu, Pengfei
Zhang, Hongyuan
Zhang, Bin
Zhou, Chunpeng
Zhou, Zheng
Zhou, Shucheng
Xie, Shuo
Zhu, Yun
Ma, Hao
Wei, Tao
Zhou, Pan
Chen, Wei
contents Traditional workflow-based agents exhibit limited intelligence when addressing real-world problems requiring tool invocation. Tool-integrated reasoning (TIR) agents capable of autonomous reasoning and tool invocation are rapidly emerging as a powerful approach for complex decision-making tasks involving multi-step interactions with external environments. In this work, we introduce MindWatcher, a TIR agent integrating interleaved thinking and multimodal chain-of-thought (CoT) reasoning. MindWatcher can autonomously decide whether and how to invoke diverse tools and coordinate their use, without relying on human prompts or workflows. The interleaved thinking paradigm enables the model to switch between thinking and tool calling at any intermediate stage, while its multimodal CoT capability allows manipulation of images during reasoning to yield more precise search results. We implement automated data auditing and evaluation pipelines, complemented by manually curated high-quality datasets for training, and we construct a benchmark, called MindWatcher-Evaluate Bench (MWE-Bench), to evaluate its performance. MindWatcher is equipped with a comprehensive suite of auxiliary reasoning tools, enabling it to address broad-domain multimodal problems. A large-scale, high-quality local image retrieval database, covering eight categories including cars, animals, and plants, endows model with robust object recognition despite its small size. Finally, we design a more efficient training infrastructure for MindWatcher, enhancing training speed and hardware utilization. Experiments not only demonstrate that MindWatcher matches or exceeds the performance of larger or more recent models through superior tool invocation, but also uncover critical insights for agent training, such as the genetic inheritance phenomenon in agentic RL.
format Preprint
id arxiv_https___arxiv_org_abs_2512_23412
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle MindWatcher: Toward Smarter Multimodal Tool-Integrated Reasoning
Chen, Jiawei
Shen, Xintian
Zheng, Lihao
Shao, Zhenwei
Cui, Handong
Du, Chaoqun
Gong, Li
Gu, Feng
Hao, Xuefeng
He, Wei
He, Jiabang
Hu, Yi
Huang, Bin
Li, Shanshan
Li, Qizhen
Luo, Jing
Liu, Zide
Liu, Xiaobo
Mao, Ning
Mu, Lifu
Pan, Xuhao
Qu, Zhiheng
Ren, Chang
Rao, Xudong
Sun, Haoyi
Wang, Qian
Wang, Shuai
Wang, Zhichao
Wang, Wei
Wen, Lian
Zhan, Jiqing
Yang, Hongfu
Yang, Sheng
Yang, Jiajun
Yu, Pengfei
Zhang, Hongyuan
Zhang, Bin
Zhou, Chunpeng
Zhou, Zheng
Zhou, Shucheng
Xie, Shuo
Zhu, Yun
Ma, Hao
Wei, Tao
Zhou, Pan
Chen, Wei
Artificial Intelligence
Traditional workflow-based agents exhibit limited intelligence when addressing real-world problems requiring tool invocation. Tool-integrated reasoning (TIR) agents capable of autonomous reasoning and tool invocation are rapidly emerging as a powerful approach for complex decision-making tasks involving multi-step interactions with external environments. In this work, we introduce MindWatcher, a TIR agent integrating interleaved thinking and multimodal chain-of-thought (CoT) reasoning. MindWatcher can autonomously decide whether and how to invoke diverse tools and coordinate their use, without relying on human prompts or workflows. The interleaved thinking paradigm enables the model to switch between thinking and tool calling at any intermediate stage, while its multimodal CoT capability allows manipulation of images during reasoning to yield more precise search results. We implement automated data auditing and evaluation pipelines, complemented by manually curated high-quality datasets for training, and we construct a benchmark, called MindWatcher-Evaluate Bench (MWE-Bench), to evaluate its performance. MindWatcher is equipped with a comprehensive suite of auxiliary reasoning tools, enabling it to address broad-domain multimodal problems. A large-scale, high-quality local image retrieval database, covering eight categories including cars, animals, and plants, endows model with robust object recognition despite its small size. Finally, we design a more efficient training infrastructure for MindWatcher, enhancing training speed and hardware utilization. Experiments not only demonstrate that MindWatcher matches or exceeds the performance of larger or more recent models through superior tool invocation, but also uncover critical insights for agent training, such as the genetic inheritance phenomenon in agentic RL.
title MindWatcher: Toward Smarter Multimodal Tool-Integrated Reasoning
topic Artificial Intelligence
url https://arxiv.org/abs/2512.23412