🧬 Awesome Auto Research

Tracking the systems that automate scientific research — from single-purpose code agents to full idea-to-paper pipelines

Autonomous research systems have gone from weekend experiments to NeurIPS Spotlight papers in under two years. This repository catalogues 30+ active projects across the full spectrum — lightweight literature scrapers, multi-agent experiment runners, and end-to-end systems that can take a vague research direction and output a reviewable manuscript — together with a capability comparison matrix, a pipeline map, a tool selection guide, and in-depth technical reports for the most impactful systems.

Latest Additions (2026-04-08): Tier updates across all sections (ARIS 5.8k★→🏆, Aider 43k★→🏆, Tongyi DeepResearch 18.6k★→🏆, SWE-agent 19k★→🏆), 7 new in-depth reports (Aider, ARIS, Tongyi DeepResearch, DeepResearchAgent, Idea2Paper, SciAgents, AIDE), Capability Matrix corrected to 18 rows

---

🗺️ Research Automation Landscape

Understanding where each tool fits in the research process is key to choosing the right one.

╔════════════════════════════════════════════════════════════════════════════════╗
║                     THE AUTONOMOUS RESEARCH PIPELINE                           ║
╠══════════════╦════════════════╦════════════════╦════════════╦══════════════════╣
║   DISCOVER   ║   SYNTHESIZE   ║  HYPOTHESIZE   ║   EXECUTE  ║  WRITE & REVIEW  ║
║              ║                ║                ║            ║                  ║
║  Idea2Paper  ║ STORM          ║ AI-Scientist   ║ OpenHands  ║ AI-Scientist     ║
║  SciAgents   ║ GPT Researcher ║ AI-Researcher  ║ SWE-agent  ║ Agent Lab        ║
║  ResAgent    ║ PaperQA2       ║ Agent Lab      ║ Aider      ║ AI-Researcher    ║
║              ║ OpenScholar    ║ autoresearch   ║ AIDE       ║                  ║
║              ║ DeerFlow       ║                ║            ║                  ║
╠══════════════╩════════════════╩════════════════╩════════════╩══════════════════╣
║                     ◄── FULL PIPELINE (End-to-End) ──►                         ║
║  autoresearch · AI-Scientist v1/v2 · AI-Researcher · Agent Laboratory · Biomni ║
╚════════════════════════════════════════════════════════════════════════════════╝

---

📑 Contents

• 📊 Capability Matrix
• 🚀 End-to-End Research Systems
• 🔍 Literature Review & Deep Research
• ⚗️ Experiment Automation & Code Agents
• ✍️ Idea Generation & Writing Assistants
• 📐 Benchmarks & Evaluation Suites
• 🎓 Academic Surveys & Papers
• 🧾 In-Depth Analysis Reports
• 🧭 How to Choose the Right Tool
• 🤝 Contributing
• 📈 Star History

---

📊 Capability Matrix

The Tier column groups systems by overall impact and maturity — this same tier label appears in every section table below, so you can quickly cross-reference.

Tier	System	Lit Review	Hypothesis	Code Exec	Paper Writing	Peer Review	Multimodal	Fully Local
🏆	OpenHands	❌	❌	✅	❌	❌	❌	✅
🏆	autoresearch	❌	✅	✅	❌	❌	❌	✅
🏆	DeerFlow	✅	❌	✅	✅	❌	❌	⚠️
🏆	STORM	✅	❌	❌	✅	❌	❌	⚠️
🏆	GPT Researcher	✅	❌	❌	✅	❌	❌	⚠️
🏆	SWE-agent	❌	❌	✅	❌	❌	❌	✅
🏆	deep-research	✅	❌	❌	✅	❌	❌	⚠️
🏆	AI-Scientist	✅	✅	✅	✅	✅	❌	⚠️
🏆	RD-Agent	❌	✅	✅	❌	❌	❌	⚠️
🏆	Open Deep Research	✅	❌	⚠️	✅	❌	❌	✅
🏆	PaperQA2	✅	❌	❌	❌	❌	❌	✅
🏆	MiroThinker	✅	❌	❌	✅	❌	⚠️	✅
🏆	ARIS	✅	✅	✅	✅	✅	❌	❌
🏆	Agent Laboratory	✅	✅	✅	✅	❌	❌	⚠️
🏆	AI-Scientist-v2	✅	✅	✅	✅	✅	❌	⚠️
🏆	AI-Researcher	✅	✅	✅	✅	❌	❌	✅
🌟	EvoScientist	✅	✅	✅	✅	✅	❌	⚠️
🌟	Biomni	✅	✅	✅	❌	❌	✅	⚠️

Tier legend: 🏆 Landmark — defined or significantly shaped the field · 🌟 Flagship — mature, widely adopted, strong results · 🔬 Notable — active, specialized, or emerging
Capability legend: ✅ Native · ⚠️ Partial / requires setup · ❌ Not supported

---

🚀 End-to-End Research Systems

Systems that automate the full research lifecycle: discovery → hypothesis → experiments → manuscript. The most ambitious category — each one aims to replace or augment the entire scientific process.

Tier	Project	Stars	Core Approach	Notes	Report
🏆	autoresearch Andrej Karpathy		630-line agent; reads its own training script, forms hypotheses, modifies code, runs hundreds of experiments overnight	Minimal & self-contained; seminal proof-of-concept	📄
🏆	AI-Scientist SakanaAI · 2024		Template-driven idea generation → experiment loop → LaTeX write-up → agentic peer review	First comprehensive end-to-end system; multiple ML research templates	📄
🏆	RD-Agent Microsoft Research · 2025		Dual-agent R&D automation: Research agent (ideation) + Development agent (implementation) with iterative loops	#1 on MLE-bench (30.22%); NeurIPS 2025; data-centric multi-domain framework	📄
🏆	AI-Scientist-v2 SakanaAI · 2025		BFTS (beam-search agentic tree search) + AIDE for code generation	First AI-written paper accepted through standard peer review	📄
🌟	DATAGEN starpig1129 · 2025		Multi-agent orchestration: hypothesis generation → data analysis → visualization → report generation	LangChain + LangGraph; advanced state tracking via Note Taker agent	📄
🏆	AI-Researcher HKUDS · NeurIPS 2025 Spotlight		LiteLLM multi-provider + Docker-sandboxed execution + Gradio UI	Broadest LLM compatibility; strong reproducibility focus	📄
🏆	Agent Laboratory SamuelSchmidgall · 2024		Role-specialized multi-agent: Professor → PhD Student → Reviewer	arXiv + HuggingFace integration for literature and datasets	📄
🌟	EvoScientist EvoScientist Team · 2026		Six-agent team (plan, research, code, analyze, write, review) with RL self-improvement	ICAIS 2025 Best Paper; #1 on DeepResearch Bench II; human-on-the-loop paradigm	📄
🌟	Biomni Stanford SNAP · 2025		Biomedical datalake + know-how library + sandboxed code execution	Domain-specialized for biology & medicine; multimodal inputs	📄
🔬	MedResearcher-R1 AQ-MedAI · 2025		KG-grounded multi-hop QA synthesis + trajectory generation for medical AI training	SOTA on MedBrowseComp; open 32B model + full training data released	📄
🔬	BioAgents bio-xyz · 2025		Specialized literature + analysis agents for biological sciences; state-of-the-art on BixBench	SOTA analysis agent (48.78% open-answer); configurable dual-agent backend	📄
🏆	ARIS wanshuiyin		Claude Code + MCP servers; runs overnight unattended	Cross-model review loops; Zotero + Obsidian integration	📄
🌟	Idea2Paper AgentAlphaAGI		Multi-agent + Knowledge Graph alignment for novelty checking	Semantic Scholar + arXiv grounding; idea → draft pipeline	📄

---

🔍 Literature Review & Deep Research

Systems specialized in information gathering, synthesis, and structured report generation. The entry point for most research workflows — and often the most practical category for daily use.

Tier	Project	Stars	Core Approach	Notes	Report
🏆	deep-research dzhng (Aomni) · 2025		Recursive depth/breadth search with Firecrawl + LLM extraction; <500 LoC reference scaffold	Most-forked deep-research scaffold; direct inspiration for Open Deep Research and DeerFlow	📄
🏆	STORM Stanford OVAL · NAACL 2024		Multi-perspective question asking + DSPy pipeline	Generates full Wikipedia-style articles with citations; Co-STORM for collaborative mode	📄
🏆	GPT Researcher assafelovic · 2023		Parallel web scraping agents + LangGraph orchestration	Outputs 5–6 page cited report (PDF / Docx / MD); MCP server support	📄
🏆	MiroThinker MiroMind AI · 2025		RL-trained open-source agent (30B / 235B) with 256K context + 300 tool calls	SOTA on BrowseComp (88.2 H1, 74.0 open); step-verifiable long-chain reasoning	📄
🌟	CognitiveKernel-Pro Tencent AI Lab · 2025		SFT-trained Qwen3-8B + Playwright web engine + multi-agent (web/file/main)	Outperforms RL-trained WebDancer/WebSailor on GAIA using SFT-only recipe; fully open model & data	📄
🏆	DeerFlow ByteDance · 2025		Sub-agent orchestration with persistent memory + InfoQuest + LangGraph	Uniquely combines deep research with code generation in one pipeline	📄
🔬	Deeper-Seeker HarshJ23 · 2024		Iterative research with follow-up questions + multi-step query generation + report synthesis	OSS alternative to OpenAI's Deep Research; Exa integration for web search	📄
🌟	PaperQA2 Future House · ICLR 2024		Iterative RAG over full-text PDFs using tantivy search index	Highest-accuracy Q&A from local scientific papers; outperforms Perplexity Pro	📄
🌟	OpenScholar Asai et al. · Nature 2024		Dense retrieval (Contriever) over 45M open-access papers	Outperforms PaperQA2 on scientific Q&A; evidence-grounded answers	📄
🌟	Open Deep Research LangChain · 2025		LangGraph workflow + MCP tool plugins + LangSmith tracing	Reference implementation from LangChain; highly configurable	📄
🌟	ToolUniverse Harvard Medical School · 2025		AI-Tool Interaction Protocol; 1,000+ tools (ML models, datasets, APIs, packages)	Universal LLM support (Claude, GPT, Gemini, Qwen, Deepseek); 68+ pre-built research skills	📄
🏆	Tongyi DeepResearch Alibaba NLP · 2025		RL-trained agentic LLM (30.5B, GRPO)	SOTA on long-horizon information-seeking benchmarks; open-weight model	📄
🌟	DeepResearchAgent Skywork AI		Hierarchical multi-agent + Autogenesis self-evolution	Planning agent coordinates specialized lower-level agents	📄
🔬	II-Researcher Intelligent Internet · 2025		BAML-structured LLM functions + multi-provider web search + async reflection loop	84.12% on Frames multi-hop benchmark; MCP server support; pip-installable	📄

---

⚗️ Experiment Automation & Code Agents

The "hands" of an autonomous research pipeline. These systems write, execute, debug, and iterate on code — essential when a hypothesis needs to become a running experiment.

Tier	Project	Stars	Core Approach	Notes	Report
🏆	OpenHands All-Hands-AI · 2024		Composable Python agent library; file editing + terminal + web browsing	72% on SWE-Bench Verified — best-in-class; production-ready UI	📄
🏆	SWE-agent Princeton NLP · 2024		Agent-Computer Interface (ACI) giving structured file/bash/edit access	~19% on SWE-Bench (full); widely used as research baseline	📄
🏆	Aider Aider-AI · 2023		AI pair programming in terminal with native Git integration	~18% on SWE-Bench; fastest daily iteration loop; supports 60+ models	📄
🔬	AutoGPT Significant Gravitas · 2023		Plugin-based autonomous agent platform + Forge builder framework	Historically seminal; sparked the autonomous agent movement	📄
🌟	AIDE WecoAI · 2024		Tree-search over ML solution space with iterative code refinement	ML-experiment-specific; used internally by AI-Scientist-v2	📄
🔬	AutoDidact dCaples · 2025		GRPO RL + self-generated Q&A pairs to bootstrap research-agent LLMs on custom corpora	Doubles Llama-8B accuracy in 1 hr on single RTX 4090; fully local open-source pipeline	📄

---

✍️ Idea Generation & Writing Assistants

Systems focused on the creative and communicative ends of research: surfacing novel hypotheses, structuring arguments, and drafting manuscripts.

Tier	Project	Stars	Core Approach	Notes	Report
🌟	Idea2Paper AgentAlphaAGI		Multi-agent pipeline with Knowledge Graph novelty alignment	Semantic Scholar + arXiv grounding; raw idea → structured research proposal	📄
🌟	SciAgents MIT · 2024		Multi-agent system with ontology graph for scientific reasoning	Generates multi-step reasoning chains grounded in domain ontologies	📄
🏆	ARIS wanshuiyin		Claude Code + MCP servers running overnight without supervision	Cross-model review loop; integrates Zotero, Obsidian, Kimi, DeepSeek	📄

---

📐 Benchmarks & Evaluation Suites

Principled evaluation frameworks for measuring the capabilities of autonomous research systems.

Benchmark	Maintained By	What It Measures	Link
SWE-Bench	Princeton NLP	Software engineering task resolution on real GitHub issues	github.com/princeton-nlp/SWE-bench
SWE-Bench Verified	OpenAI	Human-verified subset of SWE-Bench (cleaner signal)	openai.com/research
MLE-Bench	OpenAI	ML engineering quality on Kaggle competition tasks	github.com/openai/mle-bench
CORE-Bench	—	Computational reproducibility of published research	—
AI-Scientist Eval	SakanaAI	Paper quality via automated + human review	AI-Scientist
MLGym	Meta AI Research	13 open-ended AI research tasks (CV, NLP, RL, game theory) for benchmarking research agents	github.com/facebookresearch/MLGym · arXiv:2502.14499
DeepResearch Bench	Ayanami et al.	Comprehensive multi-domain benchmark for deep research agent quality	github.com/Ayanami0730/deep_research_bench
BixBench	bio-xyz	Biology-focused tool-use benchmark for research agents	github.com/bio-xyz/BioAgents
MedBrowseComp	AQ-MedAI	Medical knowledge synthesis via multi-hop web retrieval	github.com/AQ-MedAI/MedResearcher-R1

💡 Contributions to this section are especially welcome — if you know of additional evaluation suites for research agents, please open an issue or submit a PR.

---

🎓 Academic Surveys & Papers

Year	Title	Venue	Authors	Link
2024	The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery	arXiv	Lu et al. (SakanaAI)	arXiv:2408.06292
2024	From Copilot to Pilot: Towards AI-Driven Autonomous Scientific Research	arXiv	Guo et al.	arXiv:2409.14526
2024	Agent Laboratory: Using LLM Agents as Research Assistants	arXiv	Schmidgall et al.	arXiv:2501.04227
2024	STORM: Assisting in Writing Wikipedia-like Articles From Scratch	NAACL	Shao et al. (Stanford)	arXiv:2402.14207
2024	OpenScholar: Synthesizing Scientific Literature with Retrieval-Augmented LMs	Nature	Asai et al.	Nature
2024	PaperQA2: Accurate Scientific QA through Iterative Literature Search	ICLR	Skarlinski et al.	arXiv:2312.07559
2025	Towards Automated Research: A Survey of AI Agents for Scientific Discovery	arXiv	Various	—
2025	The AI Scientist-v2: Workshop-Level AI Research Automation	arXiv	Lu et al. (SakanaAI)	arXiv:2504.08066
2025	EvoScientist: Automated Scientific Discovery with Evolvable Multi-Agent Collaboration	ICAIS 2025 Best Paper	EvoScientist Team	—
2025	MLGym: A New Framework and Benchmark for Advancing AI Research Agents	arXiv	Nathani et al. (Meta)	arXiv:2502.14499
2025	SciAgents: Accelerating Scientific Discovery with Multi-Agent Intelligent Graph Reasoning	Advanced Materials	Buehler et al. (MIT)	—
2025	Tongyi DeepResearch: Reinforcement Learning for Deep Research Agents	arXiv	Alibaba NLP	—

---

🧾 In-Depth Analysis Reports

The reports/ folder is the core value of this repository. Each file contains a structured 10-section analysis: architecture internals, component breakdowns, benchmark context, and honest assessment of strengths and limitations.

📁 Browse all reports →

Tier	Report	System	Category	Key Topics Covered
🏆	ai-scientist.md	AI-Scientist	End-to-End	LaTeX pipeline, template-driven idea gen, agentic review loop
🏆	ai-scientist-v2.md	AI-Scientist v2	End-to-End	BFTS tree search, AIDE integration, peer review milestone
🏆	ai-researcher.md	AI-Researcher	End-to-End	LiteLLM multi-provider, Docker sandbox, NeurIPS 2025
🏆	agent-laboratory.md	Agent Laboratory	End-to-End	Role-specialized agents, arXiv + HuggingFace integration
🔬	biomni.md	Biomni	End-to-End	Biomedical datalake, know-how library, multimodal inputs
🔬	bioagents.md	BioAgents	End-to-End	Specialized literature + analysis agents, BixBench SOTA (48.78%)
🏆	storm.md	STORM	Literature	DSPy pipeline, multi-perspective QA, Co-STORM
🏆	gpt-researcher.md	GPT Researcher	Literature	Parallel scraping, LangGraph orchestration, MCP
🏆	deerflow.md	DeerFlow	Literature	ByteDance InfoQuest, sub-agent memory, code execution
🌟	paperqa2.md	PaperQA2	Literature	Iterative retrieval, tantivy indexing, ICLR results
🌟	openscholar.md	OpenScholar	Literature	45M paper index, Contriever dense retrieval, Nature paper
🌟	open-deep-research.md	Open Deep Research	Literature	LangChain MCP integration, LangSmith tracing
🏆	openhands.md	OpenHands	Code Agent	72% SWE-Bench Verified, composable agent architecture
🏆	swe-agent.md	SWE-agent	Code Agent	Agent-Computer Interface (ACI), Princeton NLP design
🔬	autogpt.md	AutoGPT	Code Agent	Historical context, Forge platform, Agent Protocol
🏆	autoresearch.md	autoresearch	End-to-End	630-line self-referential experiment loop, Karpathy design philosophy
🏆	deep-research.md	deep-research	Literature	Recursive depth/breadth scaffold, Firecrawl+Exa, TypeScript reference
🌟	cognitivekernel-pro.md	CognitiveKernel-Pro	Literature	SFT-trained Qwen3-8B, Playwright web engine, Tencent AI Lab
🏆	datagen.md	DATAGEN	End-to-End	Multi-agent hypothesis gen, data analysis pipeline, state tracking
🔬	medresearcher-r1.md	MedResearcher-R1	End-to-End	Medical KG-grounded trajectory synthesis, 32B model, MedBrowseComp SOTA
🏆	mirothinker.md	MiroThinker	Literature	RL-trained 30B/235B open models, 88.2 BrowseComp, interactive scaling
🔬	deeper-seeker.md	Deeper-Seeker	Literature	Iterative research, follow-up questions, multi-step synthesis
🔬	autodidact.md	AutoDidact	Code Agent	GRPO self-bootstrapping, Llama-8B, single-GPU research agent training
🔬	ii-researcher.md	II-Researcher	Literature	BAML structured LLM functions, 84.12% Frames, async multi-provider search
🏆	aider.md	Aider	Code Agent	AI pair programming, 60+ LLM models, SWE-Bench ~18%, Git-native commits
🏆	aris.md	ARIS	End-to-End	Claude Code + MCP overnight agent, cross-model review, Zotero + Obsidian
🏆	tongyi-deepresearch.md	Tongyi DeepResearch	Literature	RL-trained 30.5B (GRPO), SOTA long-horizon info-seeking, open-weight
🌟	deep-research-agent.md	DeepResearchAgent	Literature	Hierarchical multi-agent, Autogenesis self-evolution, Skywork AI
🌟	idea2paper.md	Idea2Paper	Idea Generation	Multi-agent + KG novelty alignment, Semantic Scholar + arXiv pipeline
🌟	sciagents.md	SciAgents	Idea Generation	Ontology graph + multi-agent reasoning, MIT Buehler lab
🌟	aide.md	AIDE	Code Agent	Tree-search over ML solution space, iterative code refinement, WecoAI

---

🧭 How to Choose the Right Tool

Answer the questions below in order — each branch ends at a concrete recommendation.

── START HERE ────────────────────────────────────────────────────────────────

 Q1: What is your end goal?
 │
 ├─ (A) Produce a full research paper / manuscript
 │       └─ go to Q2
 │
 ├─ (B) Survey a topic, synthesize literature, or generate a research report
 │       └─ go to Q5
 │
 ├─ (C) Run, debug, or automate code / ML experiments
 │       └─ go to Q8
 │
 └─ (D) Generate or refine novel research ideas
         └─ go to Q11

───────────────────────────────────────────────────────────────────────────────
 A: FULL PAPER / MANUSCRIPT
───────────────────────────────────────────────────────────────────────────────

 Q2: What research domain are you in?
 │
 ├─ General ML / Computer Science
 │       └─ go to Q3
 │
 ├─ Biomedical / Life Sciences
 │       └─ ✅  Biomni  (Stanford SNAP; biomedical datalake + know-how library)
 │
 └─ Other / interdisciplinary
         └─ go to Q3  (general systems are still useful starting points)

 Q3: How much control / human involvement do you want?
 │
 ├─ Fully autonomous — I want to set it running overnight
 │       └─ go to Q4
 │
 └─ Semi-autonomous — I want to steer hypothesis and review results
         └─ ✅  Agent Laboratory  (role-based: Professor → PhD Student → Reviewer;
                                    human can intervene at each stage)

 Q4: Do you prioritize pipeline maturity or LLM flexibility?
 │
 ├─ Mature pipeline, proven end-to-end results
 │       └─ ✅  AI-Scientist v1 / v2  (SakanaAI; produced first peer-reviewed AI paper)
 │
 └─ Broadest LLM provider support + reproducible Docker environment
         └─ ✅  AI-Researcher  (HKUDS; LiteLLM + Docker; NeurIPS 2025 Spotlight)

───────────────────────────────────────────────────────────────────────────────
 B: LITERATURE SURVEY / RESEARCH REPORT
───────────────────────────────────────────────────────────────────────────────

 Q5: Where does your source material come from?
 │
 ├─ The open web (news, blogs, general knowledge)
 │       └─ go to Q6
 │
 ├─ My own PDF collection (papers I've already downloaded)
 │       └─ ✅  PaperQA2  (iterative full-text RAG; highest accuracy on local PDFs)
 │
 └─ Academic papers at large scale (no local download needed)
         └─ ✅  OpenScholar  (45M open-access papers; Contriever dense retrieval;
                                Nature 2024; outperforms Perplexity Pro on sci Q&A)

 Q6: What output format do you need?
 │
 ├─ A structured, Wikipedia-style article with cited sections
 │       └─ ✅  STORM  (Stanford OVAL; DSPy pipeline; Co-STORM for collaboration;
                          NAACL 2024)
 │
 ├─ A concise 5–6 page factual report (PDF / Word / Markdown)
 │       └─ ✅  GPT Researcher  (parallel web agents + LangGraph; MCP support;
                                   fastest route to a cited report)
 │
 └─ A report that also includes runnable code or data analysis
         └─ go to Q7

 Q7: Do you need a production-grade, configurable pipeline?
 │
 ├─ Yes — I'm building this into a product or workflow
 │       └─ ✅  Open Deep Research  (LangChain; MCP tool plugins; LangSmith
                                       tracing; designed as a reference implementation)
 │
 └─ No — I need something working quickly out of the box
         └─ ✅  DeerFlow  (ByteDance; LangGraph + memory + code execution;
                             research + code in one pipeline)

───────────────────────────────────────────────────────────────────────────────
 C: CODE / EXPERIMENT AUTOMATION
───────────────────────────────────────────────────────────────────────────────

 Q8: What is your primary metric for choosing?
 │
 ├─ Raw benchmark performance on software engineering tasks
 │       └─ ✅  OpenHands  (72% on SWE-Bench Verified; best-in-class;
                               composable Python library + Web UI)
 │
 ├─ Structured, auditable, research-friendly interface
 │       └─ ✅  SWE-agent  (Princeton NLP; Agent-Computer Interface (ACI);
                               widely used as research baseline)
 │
 ├─ Daily pair-programming with Git integration (low overhead)
 │       └─ ✅  Aider  (terminal-native; Git-native commits; supports 60+ models)
 │
 └─ ML-experiment-specific iteration (Kaggle / benchmark tasks)
         └─ go to Q9

 Q9: Is your task similar to Kaggle-style ML competitions?
 │
 ├─ Yes
 │       └─ ✅  AIDE  (WecoAI; tree-search over solution space;
                          used internally by AI-Scientist-v2)
 │
 └─ No — I just want a pioneer framework to understand the space
         └─ ✅  AutoGPT  (historically seminal; Forge builder; broad plugin ecosystem)

───────────────────────────────────────────────────────────────────────────────
 D: NOVEL IDEA GENERATION
───────────────────────────────────────────────────────────────────────────────

 Q10: What kind of grounding do you need for the ideas?
 │
 ├─ Literature-grounded novelty checking (Semantic Scholar + arXiv KG)
 │       └─ ✅  Idea2Paper  (KG alignment; raw idea → structured proposal)
 │
 ├─ Domain ontology-based scientific reasoning
 │       └─ ✅  SciAgents  (MIT; multi-agent + ontology graphs)
 │
 ├─ Iterative critique loops against academic concept databases
 │       └─ ✅  ResearchAgent  (light-weight; good for early-stage idea exploration)
 │
 └─ Fully autonomous overnight ideation with cross-model review
         └─ ✅  ARIS  (Claude Code + MCP; runs unattended; Zotero + Obsidian)

── STILL UNSURE? ─────────────────────────────────────────────────────────────
 → Check the Capability Matrix above to compare any two systems side-by-side
 → Read the in-depth reports in reports/ for architecture and limitation details
──────────────────────────────────────────────────────────────────────────────

---

🤝 Contributing

• Add a project — open an issue or submit a PR to the appropriate section table
• Write an analysis report — see the report template, create reports/<slug>.md, update the Reports table above
• Fix outdated info — broken links, stale star counts, new benchmark scores
• Suggest new sections — open a Discussion

Please read CONTRIBUTING.md before submitting.

---

📈 Star History

Star growth of the leading research-specific tools since their respective launch dates.
AutoGPT (170k+ ⭐) is excluded from the chart to keep the research tools readable — view full comparison including AutoGPT →

---

Maintained by Peizheng Li · Licensed under MIT

If this repository helped your research, please consider giving it a ⭐