# GPT Researcher — An open-source autonomous research agent that plans, searches many sources and writes a cited report.

> Source: The Agents Index — https://theagentsindex.com/gpt-researcher (structured, researched, re-verified)
> Facts last verified: 2026-07-23

GPT Researcher is an open-source autonomous research agent. Given a topic or question, it plans a set of sub-questions, runs parallel web searches, reads and filters the sources, and writes a long-form report with inline citations — the ‘research agent’ pattern that later tools productised. Open-sourced by Assaf Elovic in 2023, it has since passed 28,600 GitHub stars (as of July 2026) and remains model- and search-agnostic: you bring your own API keys and plug in the LLM and search engine you prefer, so there is no subscription — you pay only the underlying API costs. Elovic went on to co-found Tavily — a paid search API for AI agents, and one of GPT Researcher's own supported search back-ends — with Rotem Weiss in late 2023, growing it directly out of this project; Nebius agreed to acquire Tavily in February 2026 for a reported ~$275M (Bloomberg; Nebius has not disclosed official terms), but GPT Researcher itself remains a separate, independent, free and open-source project, unaffected by that deal.

| Fact | Value |
| --- | --- |
| Website | https://gptr.dev |
| API | Yes |
| Best for | Developers and technical teams who want an open, self-hosted research agent they can embed and fully control. |
| Not for | Non-technical users who want a polished, hosted “ask and get a report” product with no setup. |

## Verdict

GPT Researcher is the reference open-source implementation of the autonomous research agent, and it holds up: give it a question and its planner-and-executor agents fan out across the web, read the sources and return a properly cited, long-form report. Its biggest strengths are openness and control — Apache-2.0 licensed, self-hosted, and agnostic about which LLM and search engine you use, so you can tune cost, quality and privacy to your needs, and even research your own local documents. The tradeoffs are the flip side of that freedom: it is a developer tool you have to install, key up and run, there is no hosted polish, and both quality and spend ride on the model and retriever you pick. Like every research agent, its citations cut hallucination without eliminating it, so reports still need a human check. Pick GPT Researcher if you want a free, hackable, self-hosted research engine you can build on; skip it if you want a no-setup, hosted product and are happy to pay for the convenience.

## Strengths and weaknesses

- ✓ Free and open-source (Apache 2.0) — self-host it and avoid subscriptions and lock-in.
- ✓ Model- and search-agnostic, so you control cost, quality and privacy.
- ✓ Outputs long, genuinely cited reports and can research your own local documents.
- ✗ It is a developer tool: you have to install and run it, bring your own API keys and manage the setup — there is no polished, sign-in-and-go hosted product.
- ✗ Report quality and cost depend entirely on the LLM and search back-end you choose, and Deep Research runs consume real API spend and several minutes each.
- ✗ Citations reduce but do not eliminate hallucination — it can still misattribute or summarise a source inaccurately, so output needs review.
- ⚠ Requires technical setup and your own API keys; it is not a turnkey hosted app.
- ⚠ Output fidelity is only as good as the model and retriever you choose.

## Key features

- **Planner–executor multi-agent pipeline** — A planning agent writes sub-questions, executor agents search and summarise sources, and a publisher agent composes the cited report.
- **Cited long-form reports** — Produces structured reports of 2,000+ words with inline citations, typically aggregating 20+ sources.
- **Deep Research mode** — A recursive mode that explores a topic across many branches for deeper, broader coverage.
- **Model- and search-agnostic** — Bring your own keys — OpenAI, Anthropic, Google, Groq or local Ollama, with Tavily, Google, Bing or DuckDuckGo search.
- **Multiple runtimes** — Use it as a Python package, a FastAPI + Next.js app or via Docker; export to PDF or Word. A companion project (gptr-mcp) adds an MCP server.

## Use cases

- **Automated research reports** — A developer asks GPT Researcher a question and gets a cited, multi-source report without manually gathering and reading sources.
- **Research over private documents** — Point it at local PDFs, spreadsheets or docs and it researches and cites across your own files, not just the web.
- **Embed research into an app** — A team runs it as a FastAPI service or MCP server so their own product can generate cited research on demand.

## Integrations

LLM providers — OpenAI, Anthropic, Google, Groq, local Ollama · Search — Tavily, Google, Bing, DuckDuckGo · MCP server (companion gptr-mcp project) + LangGraph multi-agent workflows · Runtimes — Python package, FastAPI + Next.js, Docker; PDF/Word export

## Sources

- https://github.com/assafelovic/gpt-researcher
- https://gptr.dev
- https://docs.tavily.com/examples/open-sources/gpt-researcher
- https://nebius.com/newsroom/nebius-announces-agreement-to-acquire-tavily-to-add-agentic-search-to-its-ai-cloud-platform
- https://www.pymnts.com/acquisitions/2026/nebius-acquires-tavily-to-bolster-agentic-ai-search-capabilities/
- https://github.com/assafelovic/gptr-mcp