MathModelAgent

MathModelAgent is an open-source AI agent that automates mathematical modeling competitions end to end, producing a complete, submission-ready paper. It has about 6,200 stars on GitHub and was actively developed in 2026.

Overview

MathModelAgent breaks a modeling contest into specialized roles — a modeler, a coder, and a paper writer — and lets each agent use a different LLM through LiteLLM, so any model can be plugged in. Code runs in a local Jupyter-based interpreter or in cloud sandboxes via E2B and Daytona, a RAG knowledge base (ChromaDB with reranking) supplies modeling methods and writing references, and web search through Tavily brings in real data. Human-in-the-loop checkpoints pause at key stages with six decision actions, and a four-layer fault-tolerance design keeps long runs alive. The current skills-based version adds 17 Typst paper templates for major competitions and a nine-step verification pass covering numeric consistency and compilation before delivery, and a desktop edition bundles everything so users only need to add an API key.

Key Features

• End-to-end automation: problem analysis, modeling, coding, error correction, and paper writing in one run
• Multi-agent, multi-model: separate modeler, coder, and writer agents, each can use a different LLM via LiteLLM
• Code interpretation: local Jupyter notebooks or cloud sandboxes via E2B and Daytona
• Submission-ready papers: formatted papers generated directly, with 17 Typst templates for major competitions
• Built-in knowledge base: RAG over modeling methods, code templates, and writing references
• Human-in-the-loop: pause at key stages to confirm, edit, regenerate, ask, skip, or abort
• Verification pass: automated checks for numeric consistency, leaks, and compile errors before delivery

💳 Pricing

MathModelAgent is open source and free for personal use — commercial use requires contacting the author. You supply your own model API keys, and a hosted online version is available on the official site.

Who Is It For

Students and teams entering mathematical modeling competitions such as MCM/ICM and national contests who want to compress days of modeling, coding, and typesetting into hours.