GPT Researcher

Open-source autonomous deep-research agent with cited long-form reports

Free· Free and open-source (MIT License). Costs come only from your chosen LLM provider (OpenAI, Anthropic, Google, etc.) and retriever (Tavily, Bing, SerpAPI, DuckDuckGo is free).AgentsModel-agnostic (default GPT-4o; supports Anthropic Claude, Google Gemini, Groq, Ollama, and any LiteLLM-compatible provider)

Visit website →

Best for

Engineers and research teams who want a self-hosted, model-agnostic deep-research agent they can embed in their own product or run on private infrastructure with their own API keys.

Skip if

Non-technical users who want a hosted chat UI with a subscription, or anyone needing sub-second single-fact lookups or real-time conversational search.

GPT Researcher is an open-source autonomous research agent that plans a research task, executes parallel web searches across multiple sources, validates and de-duplicates findings, and synthesizes a fully-cited long-form report. Instead of a single LLM call, it decomposes a query into sub-questions, spawns worker agents to gather evidence from each, and stitches the results together with inline citations, source lists, and configurable output formats (Markdown, PDF, DOCX, JSON). It supports two main modes: 'Deep Research' for multi-step investigations that return dense context and long reports, and 'Quick Search' for lower-latency retrieval with snippets. Under the hood it is LLM- and retriever-agnostic: you can wire it to OpenAI, Anthropic, Google Gemini, Groq, Ollama, or any LiteLLM-compatible backend, and pair it with Tavily, Bing, Google, SerpAPI, DuckDuckGo, Exa, Arxiv, or your own custom retriever. The project ships as a Python package with a FastAPI backend, a Next.js frontend, Docker images, and a Python SDK, so teams can run it as a standalone research app, embed it as a service inside another product, or import it as a library inside a larger agent pipeline. Typical workflows include generating market-research briefs, literature reviews, competitive analyses, due-diligence memos, and grounded context for downstream RAG or writing agents. Carnegie Mellon's May 2025 DeepResearchGym benchmark ranked it #1, ahead of Perplexity and OpenAI's Deep Research, and it has 25k+ GitHub stars with an active contributor community.

Editor's take

This is the reference implementation for open-source deep research. If you want Perplexity-style cited long-form output but need to control the model, the retriever, and where the data lives, GPT Researcher is the first thing to try — and the CMU benchmark result makes it hard to ignore. Budget for the token bill and a few hours of wiring.

— The AI Tool Bible editorial team

Pros

✅ Fully open-source (MIT) with no SaaS lock-in — self-host anywhere
✅ Model- and retriever-agnostic; swap OpenAI, Anthropic, Gemini, Ollama, Tavily, Bing, DuckDuckGo, etc.
✅ Produces long-form reports with inline citations and a source list, not just raw snippets
✅ Parallel sub-agent architecture makes multi-source research meaningfully faster than sequential prompting
✅ Ships as Python library, FastAPI service, and Next.js UI — easy to embed or run standalone
✅ Ranked #1 on CMU's DeepResearchGym (May 2025), ahead of Perplexity and OpenAI Deep Research
✅ Exports to Markdown, PDF, DOCX, and JSON out of the box

Cons

⚠️ Self-hosted only — no managed cloud, you handle deployment, keys, and quotas
⚠️ Token and retriever API costs add up on deep-research runs; a single long report can consume tens of thousands of tokens
⚠️ Quality is bounded by the LLM and retriever you choose; cheap combos produce shallow reports
⚠️ No built-in access control, team accounts, or audit log — you build governance yourself
⚠️ Latency is measured in minutes for deep research, not seconds — unsuitable for interactive chat

Use cases

Autonomous deep research reportsMarket research briefsCompetitive analysisLiterature reviewsDue-diligence memosGrounded context for downstream RAGMulti-source news synthesisResearch agent inside larger LLM pipelines