MathModelAgent is an open-source Python project that automates mathematical modeling tasks and generates a ready-to-submit paper using a multi-agent, multi-LLM setup with optional web search and local knowledge base features.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
MathModelAgent is an Agent designed for mathematical modeling that automatically analyzes problems, models, codes, debugs, and generates a formatted research paper ready for submission. It supports multiple agents and multiple LLMs, and provides features such as a Code Interpreter (local via Jupyter notebooks and cloud options), web search, a RAG knowledge base with local retrieval, and a Typst-based paper output. The project emphasizes end-to-end automation from problem analysis to paper acceptance, driven by a SKILL layer intended to be configured and extended without a Harness layer.
How it works
The repository describes a SKILL-driven workflow where stages (e.g., analysis, modeling, coding, plotting, writing) are orchestrated to complete a math modeling task. It supports injecting prompts per subtask, connects to various models via Litellm, and can leverage web search (via Tavily API) and a local knowledge base (ChromaDB + Rerank). It also supports human-in-the-loop decision actions (confirm, edit, regenerate, ask, skip, abort) and multiple fault-tolerance layers.
Getting started
Install & usage instructions are provided under the SKILLS section:
npx skills add jihe520/MathModelAgent --all
// claude
claude --dangerously-skip-permissions
claude: /1start-mathmodel 完成这个数学建模任务
// codex
codex --yolo
codex: $start-mathmodel 完成这个数学建模任务
/doctor: 检查环境配置
/typst-author: typst 知识
下载项目:
git clone https://github.com/jihe520/MathModelAgent.git
方案一为 Docker 部署，方案二为本地部署，方案三为自动脚本部署。各方案包含步骤与命令示例，如启动服务、访问前端/后端、配置 API Key，以及环境变量配置。
运行与部署要点（摘录命令）
docker-compose up
cd backend
pip install uv
uv sync
cd frontend
pnpm i
以及多处启动命令用于后端、前端及 Redis。
Recent releases
最新发布为 v0.0.1 MathModelAgent demo releases !!! (2025-05-05)，属于演示版本；仓库最近提交日期为 2026-07-20，说明仍有活跃更新。题为“最新 releases”列出该版本信息。
Traction
GitHub 级别指标：stars 3177（无时间段拆分）, forks 322, open_issues 27；语言 Python；许可证未列出。创建日期 2025-01-30，最近一次推送 2026-07-20。
Behind the repo
与 DeepWiki、Zread 等外部资源相关联的引用在文档中列出，提供外部协作与贡献渠道，但未提供具体公司/初创描述的链接信息。
Caveats
License 未列出（license: none listed）。项目处于“开发实验阶段”且有多处待实现/待改进的计划与 TODO 注释，文档中明确存在很多待改进项和公告性警示。