This repository curates research papers and benchmarks for Deep Research (DR) Agents, aiming to map the landscape and identify core concepts. It serves researchers and developers building advanced AI assistants capable of complex information retrieval and analysis. The project provides a structured overview of existing DR agents, their capabilities, and performance metrics.

How It Works

The project is a curated collection of academic papers and a comparative analysis of various DR agents. It categorizes agents by their architecture (static, dynamic single-agent, dynamic multi-agent), tool usage capabilities (code interpretation, data analytics, multimodal, MCP), and tuning methods (SFT, RL). The core value lies in its comprehensive tables that benchmark agents against tasks like search engine integration and question answering on datasets like GAIA and HLE.

Quick Start & Requirements

This repository is a collection of research and benchmarks, not a runnable agent. No installation or specific requirements are listed for running code.

Highlighted Details

• Comprehensive tables compare over 30 DR agents across search integration, tool use, architecture, tuning methods, and benchmarks.
• Benchmarks include QA datasets (Hotpot, 2Wiki, NQ, TQ, GPQA) and complex reasoning tasks (GAIA, HLE).
• Detailed breakdowns of agent capabilities, including API vs. browser usage and specific tool integrations.
• The project is actively maintained with a roadmap for "Version 2" and welcomes community contributions.

Maintenance & Community

The project is actively maintained by "ai-agents-2030" and welcomes contributions via issues or pull requests. Links to community channels or roadmaps are not provided.

Licensing & Compatibility

The repository itself does not appear to have a specific license attached. The research papers cited within would fall under their respective publication licenses.

Limitations & Caveats

This repository is a curated list and analysis, not an executable framework or library. It does not provide code for the agents discussed, only their comparative overview and performance data. The "Coming soon – Version 2!" indicates ongoing development and potential for future updates.