jihe520 / MathModelAgent
🤖 MathModelAgent 📐
MathModelAgent: An Agent specifically designed for mathematical modeling. It automatically completes mathematical modeling tasks and generates a full, submission-ready paper.
MathModelAgent：专为数学建模设计的 Agent，自动完成数学建模，生成一份完整的、可以直接提交的论文。
⬇️ Download the Latest Desktop Version (Recommended)
⬇️ 下载最新桌面版（推荐）
🎨 Sister Project: sci-box — A collection of SKILLS for scientific charts & flowcharts. 🎨 姊妹项目：sci-box —— 科研图表 & 流程图 SKILL 合集
🖥️ Desktop Version (Recommended Usage)
🖥️ 桌面版（推荐使用方式）
Don’t want to deal with environment setup? Download the desktop version for an out-of-the-box experience. 不想折腾环境？直接下载桌面版，开箱即用。
👉 Go to Releases to download the latest version. 👉 前往 Releases 下载最新版本
The desktop version comes with built-in Claude Code and the full suite of MathModelAgent SKILLS. No need to install Python, Node.js, or Redis, and no manual SKILL configuration required. Just install it, enter your model API Key, and start modeling. 桌面版已内置 Claude Code 与全套 MathModelAgent SKILLS，无需安装 Python / Node.js / Redis，也无需手动配置 SKILL，装好填一个模型 API Key 即可开始建模。
| System | Download File |
|---|---|
| macOS (Apple Silicon M-series) | mathmodel-<version>-arm64.dmg |
| macOS (Intel chip) | mathmodel-<version>-x64.dmg |
| Windows 64-bit | mathmodel-<version>-x64.exe |
The macOS installer is signed with a Developer ID and notarized by Apple. macOS 安装包已 Developer ID 签名并通过 Apple 公证。
Tip: Not sure which chip your Mac has? Click the top-left corner → About This Mac, and check the “Chip” section: If it shows Apple M-series, select arm64; if it shows Intel, select x64.
Tip：不确定自己的 Mac 是哪种芯片？点击左上角 → 关于本机，看「芯片」一栏：显示 Apple M 系列选 arm64，显示 Intel 选 x64。
Warning: The Windows installer is currently unsigned. You may see a Microsoft Defender SmartScreen prompt during the first installation or run. Please select “More info” → “Run anyway,” and ensure you download it from the official Releases page. Warning：Windows 安装包当前未签名，首次安装或运行时可能出现 Microsoft Defender SmartScreen 提示，请选择「更多信息」→「仍要运行」，并务必从官方 Releases 页面下载。
The application will automatically check for updates after installation (macOS supports auto-updates; Windows support will be enabled after the code signing certificate is configured). 安装后应用会自动检查更新（macOS 支持自动更新，Windows 待代码签名证书配置完成后启用）。
If you are a developer and want to deploy it yourself or contribute, please continue reading the SKILLS and usage tutorials below. 如果你是开发者，想自行部署或参与贡献，请继续阅读下方的 SKILLS 与使用教程。
🌟 Vision:
🌟 愿景：
Turn a 3-day competition into 1 hour. Automatically complete a prize-winning level modeling paper. 3 天的比赛时间变为 1 小时；自动完成一份可以获奖级别的建模论文。
✨ Key Features
✨ 功能特性
- 🔍 Automatic Analysis: Problem analysis, mathematical modeling, coding, error correction, and paper writing. 🔍 自动分析问题，数学建模，编写代码，纠正错误，撰写论文。
- 💻 Code Interpreter:
- Local Interpreter: Based on Jupyter; code is saved as a notebook for easy editing.
- Cloud Code Interpreter: E2B and Daytona. 💻 Code Interpreter：local Interpreter（基于 jupyter，代码保存为 notebook 方便再编辑）；云端 code interpreter（E2B 和 daytona）。
- 📝 Paper Formatting: Generate a well-formatted paper. 📝 生成一份编排好格式的论文。
- 🤝 Multi-agents: Modeling agent, coding agent, writing agent, etc. 🤝 multi-agents：建模手，代码手，论文手等。
- 🔄 Multi-llms: Set different, appropriate models for each agent. 🔄 multi-llms：每个 agent 设置不同的、合适的模型。
- 🤖 Model Support: Supports all models via LiteLLM. 🤖 支持所有模型：litellm。
- 💰 Low Cost: Workflow is agentless; does not rely on heavy agent frameworks. 💰 成本低：workflow agentless，不依赖 agent 框架。
- 🧩 Custom Templates: Prompt injection to set specific requirements for each subtask. 🧩 自定义模板：prompt inject 为每个 subtask 单独设置需求。
- 🌐 Web Search: Agent autonomously searches the internet for real data (Tavily API). 🌐 Web Search：Agent 自主搜索互联网获取真实数据（Tavily API）。
- 📚 RAG Knowledge Base: Retrieve modeling methods, code templates, and paper writing references from a local knowledge base (ChromaDB + Rerank). 📚 RAG 知识库：从本地知识库检索建模方法、代码模板、论文写作参考（ChromaDB + Rerank）。
- 🤝 HIL (Human-in-the-loop): Pauses at critical nodes for user approval; supports 6 decision actions (confirm / edit / regenerate / ask / skip / abort). 🤝 HIL 人机协作：关键节点暂停等待用户审批，支持 6 种决策动作（confirm / edit / regenerate / ask / skip / abort）。
- 🛡️ Four-layer Fault Tolerance: Limited retries → Fallback Hand Off → Evaluator Shadow Mode → Feedback Rerun. 🛡️ 四层容错：有限重试 → Fallback Hand Off → Evaluator Shadow Mode → Feedback Rerun。
I have hosted an online version on the platform for easy access. You are welcome to try it out: 我在平台中托管了一个在线版本，方便使用，欢迎体验： https://mathmodel.top/home