DeepResearchAgent

DeepResearchAgent is a hierarchical multi-agent system from Skywork AI for deep research tasks and general-purpose problem solving. A top-level planning agent breaks a goal into sub-tasks and coordinates specialised lower-level agents — for web research, browsing, document work and more — which execute them and report back. The framework was designed around benchmarks such as GAIA, the standard test of general AI assistants. The project has around 3,600 stars on GitHub and is MIT licensed.

Overview

Single agents struggle with long, branching tasks: context fills up and plans drift. DeepResearchAgent addresses this with a planner-executor hierarchy in which the top agent owns the overall plan while sub-agents work in their own focused contexts, returning distilled results. The system is multimodal, able to consume and produce text, images and other media through its specialised agents, and tool use is handled per-agent so each worker only carries the tools it needs. The codebase is structured for experimentation: agents, tools and models can be swapped, and the evaluation harness for GAIA-style tasks is included so researchers can measure changes properly.

Key Features

• Hierarchical planner-executor multi-agent architecture
• Specialised sub-agents for research, browsing and document tasks
• Automatic task decomposition and coordination across agents
• Multimodal inputs and outputs across agent workflows
• Designed and evaluated on the GAIA general-assistant benchmark
• Swappable models, tools and agent implementations
• MIT licensed

💳 Pricing

DeepResearchAgent is completely free and open source under the MIT license. You pay only for the LLM API calls your agents make while working.

Who Is It For

Researchers and developers building deep-research products or complex agent systems who want a working reference for hierarchical multi-agent design, plus teams evaluating agent frameworks against GAIA-style tasks.