Eigent Search (Q+)

Benchmarking Eigent's search agent and enhancing it with query processing toolkit (a suite of structured thinking tools).

Supported Preset Agent Types

Agent Type	Description
base	Base chat agent with no tools
search_only	Search-only agent using Google Search tool
eigent_search	Default eigent search agent with search, browse, note-taking, and terminal tools
eigent_search_q+	Eigent search agent with enhanced query processing toolkit

Supported Benchmarks

Benchmark Name	HuggingFace Data Path
SimpleQA	basicv8vc/SimpleQA
SimpleQA-Verified	google/simpleqa-verified
BrowseComp	smolagents/browse_comp
WebWalker	callanwu/WebWalkerQA
Musique	dgslibisey/MuSiQue
Frames	google/frames-benchmark

Get Started

1. Clone the repository:

git clone https://github.com/camel-ai/eigent_search_q_plus.git

2. Install dependencies:

cd eigent_search_q_plus
uv sync
source .venv/bin/activate

3. Set up environment variables:

Either export:

export OPENAI_API_KEY="your-openai-api-key"  # if you're using OPENAI backend models
export GOOGLE_API_KEY="your-google-api-key"  # for google search tool
export SEARCH_ENGINE_ID="your-search-engine-id"  # for google search tool

or save them in .env.

If you are using Azure, you need the following env variables:

export AZURE_OPENAI_BASE_URL="your-azure-url"
export AZURE_OPENAI_API_KEY="your-azure-api-key"

If you are using Minimax M2.5, you need the following env variables:

export MINIMAX_API_KEY="your-minimax-api-key"
export MINIMAX_BASE_URL="your-minimax-base-url"

4. Run the evaluation script (on the first five questions):

python scripts/simpleqa_eval.py -a eigent_search_q+ -n 5 

Please refer to the input parameters defined in the script. The benchmarking pipeline supports multiple error-handling mechanisms; if errors occur (e.g., failed problem IDs or other interruptions), use the --resume-from parameter to resume the process.

5. View results in the results directory.

License

This project is licensed under the Apache License 2.0.

Citation

@article{zhang2026eigentsearch,
  title={EigentSearch-Q+: Enhancing Deep Research Agents with Structured Reasoning Tools},
  author={Zhang, Boer and Wu, Mingyan and Zhou, Dongzhuoran and Zhu, Yuqicheng and Fan, Wendong and Zhang, Puzhen and Ding, Zifeng and Li, Guohao and He, Yuan},
  journal={arXiv preprint arXiv:2604.07927},
  year={2026}
}