DeepPlanner: Scaling Planning Capability for Deep Research Agents via Advantage Shaping

This repository contains the code and resources for the DeepPlanner project.

⚙️ Getting Started

To get started with DeepPlanner, follow these steps:

Package Installation

Install the required packages using pip:

pip install -r requirements.txt

Setup API Keys

Edit the configuration files ./user/tools/config.yaml to include your OpenAI API key and Serper API key.

openai_api_key: "your_openai_api_key"
serper_api_key: "your_serper_api_key"

For recording the experimental logs, set the Wandb API key in your ./user/train.sh and ./user/evaluate.sh scripts:

export WANDB_API_KEY="your_wandb_api_key"

Data Preparation

If you want to customize the dataset, please modify ./user/data_process/deepplanner_tool_agent_loop_data_process.py and run the data processing script:

python ./user/data_process/deepplanner_tool_agent_loop_data_process.py

🚀 Training

To train the DeepPlanner model, run the training script:

bash ./user/train.sh

🔬 Evaluation

To evaluate the trained model, use the evaluation script:

bash ./user/evaluate.sh

🌍 Citation

If you find this work useful in your research, please consider citing:

@misc{fan2025deepplannerscalingplanningcapability,
      title={DeepPlanner: Scaling Planning Capability for Deep Research Agents via Advantage Shaping}, 
      author={Wei Fan and Wenlin Yao and Zheng Li and Feng Yao and Xin Liu and Liang Qiu and Qingyu Yin and Yangqiu Song and Bing Yin},
      year={2025},
      eprint={2510.12979},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2510.12979}, 
}