GPT Researcher¶
external · status: active · focus: literature · discipline: general · started: 2023
Project page: https://github.com/assafelovic/gpt-researcher
Source: projects/landscape/gpt-researcher.yml
Positioning¶
An autonomous "deep research" agent that produces long-form, cited reports on any topic from web and local sources. Uses a planner + parallel execution-agent architecture inspired by Plan-and-Solve and RAG; outputs reports of 2,000+ words with smart image scraping and inline image generation.
Distinctive contribution¶
Earliest and most-adopted general-purpose deep-research agent; the planner/executor split with parallelized sub-queries set the reference pattern that several later systems adopted. Distributed as a pip package, Claude Skill, Docker image, and Colab notebook.
Evaluation scores¶
| Dimension | Score (0–3) | Note |
|---|---|---|
| Lifecycle coverage | 1 | Four pre-writing + drafting stages; no analysis, formal modeling, or review. |
| Autonomy level | 3 | Runs end-to-end from a research query; no per-step approval required. |
| Architectural transparency | 3 | Open source under Apache-2.0; planner/executor pattern documented; modular configuration. |
| Inputs supported | 2 | Query + optional local documents; multiple LLM and retrieval back-ends supported. |
| Outputs / reproducibility | 2 | Pip-installable, Docker image, Colab; outputs depend on live web state — not bitwise reproducible. |
| Internal evaluation | 1 | Demonstration-quality evaluation; no rigorous benchmark in the README at this scoring date. |
| Openness | 3 | Apache-2.0; broad distribution; active community. |
| Maturity / traction | 3 | 27k+ stars, regular releases, Claude Skill integration, large Discord community. |
| Cross-family policy | 0 | Multi-LLM-provider but no cross-family policy. |
| Runtime assurance | 1 | Planner/executor split with summary aggregation; no claim-audit harness. |
| Cross-platform portability | 2 | Multi-provider, Claude Skill, Docker, Colab, PyPI; deliberately portable. |
Scored on 2026-05-18. See the evaluation rubric.
Tags¶
Pipeline stages: rq-formulation literature-discovery literature-synthesis paper-drafting
Architectural features: multi-agent tool-use rag-knowledge-base iterative-loop
Inputs: research-query local-documents
Outputs: long-form-report citations inline-images
Data sources: web-search user-provided-documents
Knowledge sources: web-search
Limitations¶
- Optimized for general-purpose research; not tuned for academic-rigor outputs (no method/identification awareness).
- Output reproducibility depends on live web state.
- No formal evaluation against academic-quality benchmarks reported.
Related projects in this catalog¶
Also compared in¶
- Agentic AI for Scientific Discovery: A Survey (
gridach2025agenticsurvey) — Covered as a planner/executor deep-research agent.
Related references (literature catalog)¶
- Wu, J. et al. (2025). Agentic Reasoning: A Streamlined Framework for Enhancing LLM Reasoning with Agentic Tools
wu2025agenticreasoning