FS-Researcher

This is the code repo for paper FS-Researcher (currently a minimal implementation, performance not guaranteed, complete code coming soon).

Project layout

• main.py: CLI entrypoint (runs two agents: Stage 1 + Stage 2)
• agent.py: minimal ReAct-style agent + tool executor
• tools.py: tools including search_web (Google Serper), read_webpage (Jina AI Reader), and FileTools (local filesystem utilities)
• prompts.py: system prompts for Stage 1 / Stage 2
• config.py: local config (reads from environment variables first)
• requirements.txt: minimal dependencies

Installation

Create a virtual environment if you prefer, then:

pip install -r research_agent/requirements.txt

Configuration (environment variables)

• OpenAI-compatible API

  • OPENAI_API_KEY: required
  • OPENAI_API_BASE: optional, defaults to https://api.openai.com/v1
  • LLM_MODEL: optional, defaults to gpt-5

• Tools

  • SERPER_API_KEY: Google Serper key (used by search_web)
  • JINA_API_KEY: Jina Reader key (used by read_webpage, optional)

Run

From the repository root:

python -m research_agent.main --topic "your research topic" --workspace ./workspace_demo

You can also copy the entire research_agent/ folder into another project and run it from its parent directory:

python -m research_agent.main --topic "your research topic"