In the field of computer networks, innovation is crucial for both the academic and industrial sectors. As more researchers engage in this important research area, peer competition intensifies. However, traditional methods of innovation are hindered by personal biases, knowledge boundaries, time costs and resource constraints, which impede scientific breakthroughs in the network domain. In this paper, we introduce NetInno, a multi-agent cyclic framework. NetInno is based on Large Language Models (LLMs) and is capable of autonomous innovation. It utilizes feedback to further optimize results, thereby reducing the need for human intervention. Each agent employs tailored prompting strategies to enhance expertise, eliminating the need for extensive initial setup and model training. Human review ensures the professionalism of the feedback, addressing potential “hallucination” issues of LLMs. Our initial experiments and case studies validate the effectiveness and applicability of NetInno, demonstrating its potential to automate innovation through iterative frameworks and tailored strategies. We believe this represents the first step towards developing fully automated systems for research innovation in the future of networking.