社区开发者 @dvictor3579 用 MiniCPM5-1B 构建了 GMGN Local Research Agent，通过本地运行和原生工具调用连接 GMGN 数据，实现安全只读的研究工作流。该项目主打本地优先，支持私有设备端推理、低延迟且无额外 LLM API 成本，展示了紧凑开源模型在实用本地智能体中的应用。代码与模型已分别开源至 GitHub 和 HuggingFace。
A developer in our community @dvictor3579 , built GMGN Local Research Agent with MiniCPM5-1B 🥳
Instead of relying on frontier models, this project runs MiniCPM5-1B locally and connects GMGN data through native tool calling for safe, read-only research workflows.
Designed for local-first AI agents, it enables:
🧠 Native tool calling with lightweight language models
🔒 Private, on-device inference without cloud LLM dependency
⚡ Low-latency research workflows with no additional LLM API cost
🛡️ Safe read-only access for market, token, wallet, and smart money research
Built with MiniCPM5-1B, this project highlights how compact open-source models can enable efficient local AI agents for practical, real-world use cases.
来源：OpenBMB · x.com