287 ·GitHub
What is this?
Modex · MH Agent is a Windows desktop AI agent that automates the full pipeline of mathematical modeling competitions and academic research, taking a competition problem all the way to a competition-grade paper overnight with one command.
Modex · MH Agent is a Windows desktop AI agent that automates the full pipeline of mathematical modeling competitions and academic research, taking a competition problem all the way to a competition-grade paper overnight with one command.Is this repository the actual app, or just documentation?This GitHub repo is only a product showcase and issue-feedback channel — it contains no source code at all; the real app is a downloadable Windows desktop program distributed from the official website.
This GitHub repo is only a product showcase and issue-feedback channel — it contains no source code at all; the real app is a downloadable Windows desktop program distributed from the official website.How do I get it running?Download the installer from the official site (mingheng.xin) and launch it — Python, Node, Git, TeXLive, MiKTeX, Pandoc, draw.io and the Claude CLI are all bundled, so no manual environment setup is needed. Then open Settings, enter your executor and reviewer model API keys, test the connection, pick a competition or research template, fill in the parameters, and start the run.
Download the installer from the official site (mingheng.xin) and launch it — Python, Node, Git, TeXLive, MiKTeX, Pandoc, draw.io and the Claude CLI are all bundled, so no manual environment setup is needed. Then open Settings, enter your executor and reviewer model API keys, test the connection, pick a competition or research template, fill in the parameters, and start the run.Do I need my own Claude and reviewer API keys?Yes — the software itself ships no model quota. You must supply your own Claude API key as the executor plus an OpenAI-compatible API key (GPT, GLM, MiniMax, Kimi, DeepSeek, etc.) as the reviewer, and all model calls are billed to those accounts.
Yes — the software itself ships no model quota. You must supply your own Claude API key as the executor plus an OpenAI-compatible API key (GPT, GLM, MiniMax, Kimi, DeepSeek, etc.) as the reviewer, and all model calls are billed to those accounts.Does Modex need an internet connection, and can I trust its output as-is?Yes, it needs a stable connection at all times for literature search, model calls and citation checking (users in mainland China may need a proxy for some services), and the README explicitly warns that AI-generated content, data and citations must be manually verified before use.
Yes, it needs a stable connection at all times for literature search, model calls and citation checking (users in mainland China may need a proxy for some services), and the README explicitly warns that AI-generated content, data and citations must be manually verified before use.Which math modeling competitions does it cover?It has full pipelines for China's CUMCM national contest, the US MCM/ICM, Huawei Cup, MathorCup, APMCM and open statistical-modeling contests, each with its own paper template, page-length target and formatting rules.
It has full pipelines for China's CUMCM national contest, the US MCM/ICM, Huawei Cup, MathorCup, APMCM and open statistical-modeling contests, each with its own paper template, page-length target and formatting rules.What is the cross-model adversarial review it mentions?Claude acts as the executor while a separate model (GPT, GLM, MiniMax, etc.) acts as the reviewer, so no model grades its own work — the idea is that cross-model review is harder to game than a single model checking itself.
Claude acts as the executor while a separate model (GPT, GLM, MiniMax, etc.) acts as the reviewer, so no model grades its own work — the idea is that cross-model review is harder to game than a single model checking itself.What are the 90+ Skills it ships with?Each Skill is a plain SKILL.md markdown file covering a specific task — literature review, problem analysis, modeling, chart generation, LaTeX/Mermaid/draw.io diagrams, thesis writing, grant proposals — and they can be freely mixed or chained into a pipeline by any LLM.
Each Skill is a plain SKILL.md markdown file covering a specific task — literature review, problem analysis, modeling, chart generation, LaTeX/Mermaid/draw.io diagrams, thesis writing, grant proposals — and they can be freely mixed or chained into a pipeline by any LLM.Can it be used for general academic research, not just competitions?Yes, besides competition papers it covers an end-to-end research workflow: literature review, idea discovery, novelty verification, experiment deployment, automated peer review, and writing a submission-ready PDF.
Yes, besides competition papers it covers an end-to-end research workflow: literature review, idea discovery, novelty verification, experiment deployment, automated peer review, and writing a submission-ready PDF.What are similar projects?PaperSpine, FigGenie-paper-diagram-skill, cumcm-step-review, codex-claude-academic-skills, evenfire, claude-scholar, vibe-coding-cn, ordewell
PaperSpine, FigGenie-paper-diagram-skill, cumcm-step-review, codex-claude-academic-skills, evenfire, claude-scholar, vibe-coding-cn, ordewellWhich topics is it in?paper-writing, academic-writing, skill-md, latex, mathematics, document-processing, workflow-automation, windows-app, websocket, academic-research, openai-compatible, agent-orchestration
paper-writing, academic-writing, skill-md, latex, mathematics, document-processing, workflow-automation, windows-app, websocket, academic-research, openai-compatible, agent-orchestration