gpt-researcher

GPT Researcher is an open-source autonomous research agent designed to conduct in-depth investigations on any given topic by leveraging both web and local document sources. It generates comprehensive, factual, and citation-backed research reports, aiming to reduce misinformation and bias through parallelized agent work and aggregation of multiple sources. The system follows a planner-executor-publisher architecture: the planner formulates research questions, execution agents gather data via web scraping (including JavaScript-enabled crawling) and local file parsing (PDF, Word, Excel, etc.), and the publisher compiles findings into structured reports exceeding 2,000 words. Key features include AI-generated inline images using Google Gemini models, support for over 20 aggregated sources per report, export to PDF/Word formats, and optional integration with LangSmith or Monocle for observability. It can be installed as a Python package, Docker container, or Claude Skill, enabling seamless use within conversational AI interfaces. Advanced capabilities include Deep Research — a recursive, tree-based exploration workflow — and Multi-Agent Assistant frameworks built with LangGraph and AG2, inspired by STORM methodology. Users can customize retrieval methods (e.g., Tavily + MCP for GitHub/database access), enable local document research via DOC_PATH, and configure frontend options (lightweight FastAPI or production-grade NextJS). The project emphasizes transparency, community collaboration, and academic experimentation under Apache 2 license, while acknowledging its experimental nature and limitations regarding absolute objectivity.