---
dataset_info:
  features:
  - name: agent_id
    dtype: string
  - name: platform_name
    dtype: string
  - name: agent_name
    dtype: string
  - name: agent_description
    dtype: string
  - name: agent_category
    dtype: string
  - name: agent_usage
    dtype: string
  - name: usage_example
    dtype: float64
  - name: agent_url
    dtype: string
  - name: agent_accessibility
    dtype: string
  - name: agent_pricing
    dtype: string
  - name: base_model
    dtype: string
  - name: update_time
    dtype: string
  - name: misc
    dtype: string
  - name: platform_id
    dtype: string
  splits:
  - name: agents
    num_bytes: 14529673
    num_examples: 9759
  download_size: 6044862
  dataset_size: 14529673
configs:
- config_name: default
  data_files:
  - split: agents
    path: data/train-*
task_categories:
- text-retrieval
---

# AgentSearchBench Agents

**AgentSearchBench** is a large-scale benchmark for AI agent search, built from nearly 10,000 real-world agents sourced from the [GPT Store](https://chatgpt.com/gpts), [Google Cloud Marketplace](https://cloud.google.com/marketplace), and [AgentAI Platform](https://agent.ai/).

📄 [Paper](https://huggingface.co/papers/2604.22436) • 🌐 [Project Page](https://bingo-w.github.io/AgentSearchBench) • 💻 [Codebase](https://github.com/Bingo-W/AgentSearchBench)

---

## Overview

This repository contains **AgentBase**, the agent corpus underlying AgentSearchBench. It comprises 9,759 real-world AI agents crawled from public platforms, capturing practical challenges such as capability overlap and inconsistent documentation that reflect the complexity of real agent ecosystems.

Each agent entry includes metadata such as name, description, capabilities, and platform of origin, forming the candidate pool against which benchmark tasks are evaluated.

---

## Dataset Statistics

| Platform | Description |
|----------|-------------|
| GPT Store | Custom GPT agents from OpenAI's public marketplace |
| Google Cloud Marketplace | Cloud-native agents and tools |
| AgentAI Platform | General-purpose agents from agent.ai |

| Total Agents |
|---|
| 9,759 |

---

## Data Fields

- `agent_id`: Unique identifier for the agent.
- `platform_name`: Name of the platform where the agent is hosted.
- `agent_name`: Display name of the agent.
- `agent_description`: Text description of the agent's purpose and capabilities.
- `agent_category`: Category or domain the agent belongs to.
- `agent_usage`: Instructions or notes on how to use the agent.
- `usage_example`: Example input or interaction demonstrating the agent's functionality.
- `agent_url`: URL linking to the agent's page on its host platform.
- `agent_accessibility`: Accessibility status of the agent, e.g. public or restricted.
- `agent_pricing`: Pricing model or cost information associated with using the agent.
- `base_model`: Underlying language model powering the agent.
- `update_time`: Timestamp of the agent's last update.
- `misc`: Miscellaneous metadata not captured by other fields.

## Usage

```python
from datasets import load_dataset

ds = load_dataset("AgentSearch/AgentSearchBench-Agents")
```

---

## Related Datasets

| Dataset | Description |
|---------|-------------|
| [AgentSearchBench-Tasks](https://huggingface.co/datasets/AgentSearch/AgentSearchBench-Tasks) | Benchmark tasks: single-agent queries, multi-agent queries, and task descriptions |
| [AgentSearchBench-Responses](https://huggingface.co/datasets/AgentSearch/AgentSearchBench-Responses) | 60K+ raw agent execution responses from the validation set |

---

## Citation

```bibtex
@misc{wu2026agentsearchbench,
      title={AgentSearchBench: A Benchmark for AI Agent Search in the Wild}, 
      author={Bin Wu and Arastun Mammadli and Xiaoyu Zhang and Emine Yilmaz},
      year={2026},
      eprint={2604.22436},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
}
```