An army of LLM interns that actually cite their sources

What it does GPT Researcher tackles a research query by delegating it to parallel agents: a planner breaks the topic into questions, execution agents scrape answers from the web or local files, and a publisher assembles a cited, long-form report. It can also recurse into subtopics in a tree-like “Deep Research” mode, and pull from private data sources via MCP.

The interesting bit The project treats research as an orchestration problem rather than a single prompt. By parallelizing agents and maintaining memory across the pipeline, it attempts to beat context-window limits and hallucination with brute-force coordination instead of a bigger model. The Deep Research mode even quotes a price—about $0.40 and five minutes per run on o3-mini high—which is refreshingly concrete for an AI tool.

Key highlights

• Planner / execution / publisher architecture with parallelized agents and source tracking.
• Deep Research: recursive, tree-like exploration with configurable depth and breadth.
• Hybrid retrieval: web search plus MCP connectors for GitHub, databases, and custom APIs.
• Local document ingestion: PDF, Word, Excel, CSV, Markdown, and PowerPoint.
• Optional inline image generation via Google Gemini with dark-mode styling.

Verdict Worth a look if you regularly compile research briefs, competitive analysis, or literature reviews and want a reproducible pipeline rather than a chat session. Skip it if you expect fully offline operation or are allergic to wiring up API keys for search and inference.