GPT Researcher: The Open Deep Research Agent Reviewed

Key highlights:

• 29K+ GitHub stars, Apache-2.0, Python, created May 2023 and still actively maintained (last push July 2026)

• Planner + executor architecture: one agent decomposes your query into sub-questions, parallel agents gather sources, a publisher aggregates the report

• Provider-agnostic: works with OpenAI, Anthropic, Google, Groq, Ollama, and any OpenAI-compatible endpoint

• Real citations: aggregates 20+ sources per report and tracks each claim back to its origin

• Deep Research mode: recursive tree-of-thought exploration that trades cost and time for depth

Deep research has become the flagship feature of every major AI lab - OpenAI, Google, Anthropic, and Perplexity all ship a "research mode" that spends minutes instead of seconds to produce a long, cited answer. GPT Researcher is the open-source project that got there first, and it remains the most popular self-hosted option.

Built by Assaf Elovic, GPT Researcher is described in its own words as "the first open deep research agent designed for both web and local research on any given task." Instead of a single LLM call, it orchestrates a small team of agents: a planner that reasons about what to investigate, execution agents that go find the information, and a publisher that assembles everything into a coherent, referenced report.

The design is explicitly inspired by two research papers - Plan-and-Solve prompting and Retrieval-Augmented Generation (RAG) - and it targets the failure modes that plague naive LLM research: hallucination from outdated training data, token limits that cap report length, and bias from a narrow set of sources.

The core idea is a split between planner and execution agents. Here is the pipeline it runs for every query:

1. Create a task-specific agent tuned to your research query.

2. Generate research questions - a set of sub-questions that together form an objective view of the topic.

The fastest way to embed research into your own app is the Python package:

export TAVILY_API_KEY="tvly-..." # default web retriever

Tavily is the default search retriever, but you are not locked in - GPT Researcher supports a long list of retrievers including DuckDuckGo (no key), Google, Bing, SearchAPI, SerpAPI, Exa, and local document search.

This is a condensed version. The full article includes architecture diagrams, code examples, comparison tables, and practical use cases.

👉 Read the complete analysis on andrew.ooo

Topics: ai agent, deep research, gpt-researcher, open-source, rag