Open-source Deep Research
About AutoSearch
AutoSearch started in 2026 as MIT-licensed, open-source deep research infrastructure for agent hosts. The project focuses on MCP-native retrieval across 40 channels, 10+ Chinese sources, and an LLM-decoupled architecture that keeps model choice in the host.
Roadmap
- More Chinese channels for technical, finance, and product research.
- Evals dashboard for source quality, citation coverage, and channel reliability.
- Agent-host SDK improvements for smoother MCP-native integration.
Project
Built for developers who want deep research tools that stay portable across hosts, models, and source ecosystems.
Founder
0xmariowu
Background
0xmariowu has been shipping developer tooling for AI agents since the MCP standard appeared. The work on AutoSearch grew out of frustration with deep-research tools that hard-coded a single LLM, ignored Chinese sources, and treated retrieval like a finished problem. The goal: portable, evidence-first research that any agent host can plug into.
Motivation
Existing deep-research products bind reasoning, retrieval, and presentation into one closed loop. That rules out half the world's information ecosystem and locks you into one vendor's roadmap. AutoSearch separates the layers — host keeps the LLM, AutoSearch handles retrieval — so research stays open and inspectable as the model layer evolves.