đ¤ MathModelAgent đ
An agent designed for mathematical modeling
Automatically complete mathematical modeling and generate a ready-to-submit paper.
## đ Vision
Turn 3 days of competition into 1 hour
Automatically generate an award-level modeling paper
## ⨠Features
- đ Automatic problem analysis, mathematical modeling, code writing, error correction, and paper writing
- đģ Code Interpreter
- Local Interpreter: Based on Jupyter, code saved as notebook for easy editing
- Cloud Code Interpreter: [E2B](https://e2b.dev/) and [daytona](https://app.daytona.io/)
- đ Generate a well-formatted paper
- đ¤ Multi-agents: modeling expert, coding expert, paper expert, etc.
- đ Multi-LLMs: Different models for each agent
- đ¤ Support for all models: [litellm](https://docs.litellm.ai/docs/providers)
- đ° Low cost: workflow agentless, no dependency on agent framework
- 𧊠Custom templates: prompt inject for setting requirements for each subtask separately
## đ Future Plans
- [x] Add and complete webui, cli
- [ ] Comprehensive tutorials and documentation
- [ ] Provide web service
- [ ] English support (MCM/ICM)
- [ ] LaTeX template integration
- [ ] Vision model integration
- [x] Proper citation implementation
- [x] More test cases
- [x] Docker deployment
- [ ] Human in loop: User interaction (model selection, @agent rewriting, etc.)
- [ ] Feedback: evaluate the result and modify
- [x] Cloud integration for code interpreter (e.g., e2b providers)
- [ ] Multi-language: R, Matlab
- [ ] Drawing: napki, draw.io, plantuml, svg, mermaid.js
- [ ] Add benchmark
- [ ] Web search tool
- [ ] RAG knowledge base
- [ ] A2A hand off: Code expert reflects on errors multiple times, hands off to smarter model agent
## Video Demo
> [!CAUTION]
> The project is in experimental development stage, with many areas needing improvement and optimization. I (the project author) am busy but will update when time permits.
> Contributions are welcome.
For case references, check the [demo](./demo/) folder.
**If you have good cases, please submit a PR to this directory**
## đ Usage Guide
Three deployment options are available, choose the one that suits you best:
1. Docker
2. Local deployment
3. Automated script deployment
> If you want to run the CLI version, switch to the [master](https://github.com/jihe520/MathModelAgent/tree/master) branch. It's easier to deploy, but will not be updated in the future.
### đŗ Option 1: Docker Deployment (Recommended: Simplest)
1. Configure Environment Variables
```bash
cp backend/.env.dev.example backend/.env.dev
cp frontend/.env.example frontend/.env.development
```
Fill in the configuration in:
- backend/.env.dev
- frontend/.env.development
2. Start Services
```bash
docker-compose up -d
```
3. Access
You can now access:
- Frontend interface: http://localhost:5173
- Backend API: http://localhost:8000
### đģ Option 2: Local Deployment
> Make sure Python, Nodejs, and **Redis** are installed on your computer
1. Configure Environment Variables
Copy `/backend/.env.dev.example` to `/backend/.env.dev` (remove the `.example` suffix)
**Configure Environment Variables**
It is recommended to use models with strong capabilities and large parameter counts.
Copy `/frontend/.env.example` to `/frontend/.env.development` (remove the `.example` suffix)
2. Install Dependencies
Clone the project
```bash
git clone https://github.com/jihe520/MathModelAgent.git
```
Start backend
*Start Redis*
```bash
cd backend
pip install uv # Recommended: use uv to manage python projects
uv sync # Install dependencies
# Start backend
# Activate Python virtual environment
source .venv/bin/activate # MacOS or Linux
venv\Scripts\activate.bat # Windows
# Run this command for MacOS or Linux
ENV=DEV uvicorn app.main:app --host 0.0.0.0 --port 8000 --ws-ping-interval 60 --ws-ping-timeout 120 --reload
# Run this command for Windows
set ENV=DEV ; uvicorn app.main:app --host 0.0.0.0 --port 8000 --ws-ping-interval 60 --ws-ping-timeout 120
```
Start frontend
```bash
cd frontend
npm install -g pnpm
pnpm i # Make sure pnpm is installed
pnpm run dev
```
[Tutorial](./docs/md/tutorial.md)
Results and outputs are generated in the `backend/project/work_dir/xxx/*` directory:
- notebook.ipynb: code generated during execution
- res.md: final results in markdown format
### đ Option 3: Automated Script Deployment (Community Contribution)
Need an automatic deployment script?
[mmaAutoSetupRun](https://github.com/Fitia-UCAS/mmaAutoSetupRun)
Need to customize prompt templates?
Prompt Inject: [prompt](./backend/app/config/md_template.toml)
## đ¤ Contribution & Development
[DeepWiki](https://deepwiki.com/jihe520/MathModelAgent)
- The project is in **experimental development stage** (updated when I have time), with frequent changes and some bugs being fixed.
- Everyone is welcome to participate and make the project better.
- PRs and issues are very welcome.
- For requirements, refer to Future Plans.
After cloning the project, install the **Todo Tree** plugin to view all todo locations in the code.
`.cursor/*` contains overall architecture, rules, and mcp for easier development.
## đ License
Free for personal use. For commercial use, please contact me (the author).
[License](./docs/md/License.md)
## đ Reference
Thanks to the following projects:
- [OpenCodeInterpreter](https://github.com/OpenCodeInterpreter/OpenCodeInterpreter/tree/main)
- [TaskWeaver](https://github.com/microsoft/TaskWeaver)
- [Code-Interpreter](https://github.com/MrGreyfun/Local-Code-Interpreter/tree/main)
- [Latex](https://github.com/Veni222987/MathModelingLatexTemplate/tree/main)
- [Agent Laboratory](https://github.com/SamuelSchmidgall/AgentLaboratory)
## Others
### đ Sponsor
[Buy Me a Coffee](./docs/sponser.md)
Thanks to sponsors:
[danmo-tyc](https://github.com/danmo-tyc)
### đĨ GROUP
For questions, join the group
[QQ Group: 699970403](http://qm.qq.com/cgi-bin/qm/qr?_wv=1027&k=rFKquDTSxKcWpEhRgpJD-dPhTtqLwJ9r&authKey=xYKvCFG5My4uYZTbIIoV5MIPQedW7hYzf0%2Fbs4EUZ100UegQWcQ8xEEgTczHsyU6&noverify=0&group_code=699970403)
