An open, community-driven repository for AI-assisted scientific research
Curating agents, skills, workflows, tools & best practices across the entire research lifecycle
🇨🇳 中文版 | 🇬🇧 English
🧪 End-to-End • 🔭 Topic Discovery • 📚 Literature • 🧩 Method Design • ⚗️ Experiments • 📊 Visualization • ✍️ Writing • 📦 Reproduction • 📡 Dissemination • 🤝 Contribute
We focus on AI-assisted capabilities that are reusable across research workflows, verifiable by peers, evolvable over time, and state-of-the-art in performance.
This repository grew out of the "Doing Research with Agents" seminar series. We aim to consolidate speakers' hands-on experience, audience feedback, open-source projects, and reusable workflows into a community-maintained knowledge base.
The table below shows our 9-stage decomposition of the research lifecycle. Each stage lists "typical questions" and "types of AI-assisted components that can be distilled." The main body expands each stage into a curated project table.
💡 See something missing? Just add a row to the relevant table!
| Stage | Typical Questions | Scope |
|---|---|---|
| 🔄 0. End-to-End | Spans multiple stages below | — |
| 🔭 1. Topic Discovery & Problem Definition | What's happening in this field? What problems are worth pursuing? | trend scanner, paper radar, venue tracker; idea generator, novelty checker, hypothesis workflow |
| 📚 2. Literature Research | How to find, read, compare, and synthesize related work? | literature review workflow, paper reading skill, citation graph agent |
| 🧩 3. Method Design | How to design the approach, experiments, and evaluation metrics? | experiment design skill, ablation planner, protocol checker |
| ⚗️ 4. Experiment Execution & Analysis | How to code, run experiments, and log failures? Are results reliable? | experiment runner, statistical analysis skill, failure analysis workflow, robustness checker |
| 📊 5. Visualization | Do the figures clearly communicate the scientific question? | figure generation agent, visualization critique skill |
| ✍️ 6. Paper Writing | How to structure the paper, manage citations, supplement experiments? | paper writing workflow, citation verifier, rebuttal assistant |
| 📦 7. Reproduction & Release | How to enable others to reproduce and use your work? | artifact packaging workflow, model card, data card, reproducibility checklist |
| 📡 8. Dissemination & Impact | How to track impact after publication? How to build academic influence? | impact analysis tool, social summary skill, citation monitor |
End-to-end systems from idea to paper. Agents spanning three or more stages go here; single-stage tools belong in their respective section.
| Project | Description | Type | Stars | Link | Demo | Paper |
|---|---|---|---|---|---|---|
| AI-Scientist | End-to-end automated scientific discovery: idea generation, experiments, paper writing, and peer review | agent | ⭐ 14k | GitHub | - | Nature 2024 |
| AI-Scientist-v2 | Agentic tree search for automated research, template-free, targeting general ML exploration | agent | ⭐ 6.6k | GitHub | - | arXiv 2025 |
| EvoScientist | Multi-agent AI scientist with persistent memory, skill evolution, and end-to-end research collaboration | agent | ⭐ 3.6k | GitHub | Demo | arXiv 2025 |
| nature-skills | Skills for Nature-grade academic writing and scientific figure creation | skill | ⭐ 20.9k | GitHub · ModelScope Skills | - | - |
| AutoResearchClaw | Autonomous, self-evolving multi-stage research pipeline from idea to paper artifacts | agent | ⭐ 13.5k | GitHub | Demo | arXiv 2025 |
| autoresearch | Andrej Karpathy's autonomous ML research agent running experiments on a single GPU | agent | ⭐ 87.4k | GitHub | - | - |
| Auto Claude Code Research in Sleep | Automated Claude Code research workflow for unattended experiment execution | skill | ⭐ 12.3k | GitHub | - | - |
| AgentLaboratory | End-to-end autonomous research workflow with LLM-powered specialized agents for literature review through report writing | agent | ⭐ 5.7k | GitHub | - | - |
| Aether | Open-source AI research environment with unified web & desktop experience, powered by research-focused agents and skills | agent/app | ⭐ 65 | GitHub | - | - |
| EurekAgent | Environment-engineered autonomous research system for metric-driven tasks, coordinating Claude Code sessions to propose, implement, evaluate, and iterate solutions | agent | ⭐ 55 | GitHub | - | arXiv 2026 |
| InternAgent | Unified agentic framework for long-horizon autonomous scientific discovery, covering hypothesis generation, automated experimentation, paper reproduction, memory, and Deep Research | agent | ⭐ updating | GitHub | Website | arXiv 2026 |
| Tashan Research Skills | 16 research skills built by the Tashan team at UCAS, organized into literature evidence, ideation, expression, collaboration memory, and tool evaluation; each skill ships with its own scripts, templates, and tests | skill | ⭐ updating | GitHub | - | - |
Trend tracking, research inspiration, novelty verification, hypothesis generation. When overlapping with literature search, the deciding factor is: "Does it output a viable research question?"
| Project | Description | Type | Stars | Link | Demo | Paper |
|---|---|---|---|---|---|---|
| SciAgentsDiscovery | MIT open-source scientific discovery system combining knowledge graphs and multi-agent collaboration for cross-domain hypothesis generation | agent | ⭐ 615 | GitHub | - | arXiv 2024 |
| AutoDiscovery | AllenAI open-ended scientific discovery framework using Bayesian Surprise to find testable hypotheses from data | benchmark/workflow | ⭐ 185 | GitHub | - | NeurIPS 2024 |
Literature search, RAG Q&A, automated survey generation, citation graph analysis. Pure writing/polishing tools go in Stage 6.
| Project | Description | Type | Stars | Link | Demo | Paper |
|---|---|---|---|---|---|---|
| STORM | Stanford's open-source knowledge curation system generating cited reports via multi-perspective question generation | agent | ⭐ 28.5k | GitHub | Demo | NAACL 2024 |
| paperseek | A researcher-oriented literature discovery tool supporting natural language queries with iterative search expansion | agent/skill | ⭐ 36 | GitHub | paperseek.xyz | - |
| OpenAlex Search Skill | OpenAlex is an open global scholarly graph covering works, authors, institutions, venues, and citations. This Codex skill packages OpenAlex works search into a reusable command-line workflow; recommended defaults are title + abstract Boolean search, include_xpac=true, and stemming enabled, with optional customization through semantic mode, citation sorting, year filters, XPAC/stemming switches, and JSON output. |
skill | ⭐ 0 | GitHub | Skill | - |
| arXiv Browser Research Skill | Codex/browser-use skill for polite arXiv browser fallback when OpenAlex, Semantic Scholar, or the arXiv API are rate limited or incomplete; extracts structured titles, authors, abstracts, arXiv IDs, and PDF links from arXiv result pages. | skill | ⭐ updating | GitHub | Skill | - |
| Lune | MCP server providing agentic search over top-tier papers, with grounding for both academic literature and research best practices | agent/tool | ⭐ 2 | GitHub | Demo | - |
| PaperQA2 | High-accuracy RAG system for scientific papers, producing evidence-grounded answers with citations | python pkg | ⭐ 8.7k | GitHub | - | - |
| OpenScholar | Retrieval-augmented scientific literature synthesis system for generating scholarly answers grounded in open corpora | agent/model | ⭐ 1.5k | GitHub | - | arXiv 2024 |
| paper-search-mcp | MCP/CLI/Skill for agent-facing paper search across arXiv, PubMed, bioRxiv, Semantic Scholar, OpenAlex, and more | tool/skill | ⭐ 1.9k | GitHub | - | - |
| Zotero-GPT | AI literature-reading plugin inside Zotero, supporting summarization, Q&A, tagging, and note assistance | plugin | ⭐ 7.2k | GitHub | - | - |
Experiment planning, evaluation metric design, ablation planning, protocol checking. Pure coding and experiment running go in Stage 4.
| Project | Description | Type | Stars | Link | Demo | Paper |
|---|---|---|---|---|---|---|
| Curie | Automated and rigorous scientific experimentation agent, spanning hypothesis clarification, execution, analysis, and reporting | agent/workflow | ⭐ 363 | GitHub | - | arXiv 2025 |
Coding, experiment running, statistical analysis, failure analysis, robustness checking. Dataset construction goes in Stage 2/3; visualization in Stage 5.
| Project | Description | Type | Stars | Link | Demo | Paper |
|---|---|---|---|---|---|---|
| RD-Agent | Automates high-value R&D processes for data and models — letting AI drive data-driven AI | agent | ⭐ 13.5k | GitHub | - | arXiv 2025 |
| EurekAgent | Execution environment for metric-driven research tasks, supporting Claude Code sessions for implementation, Docker-isolated evaluation, logging, and iterative optimization | agent | ⭐ 55 | GitHub | - | arXiv 2026 |
Publication-quality figures, schematic generation, data dashboards. Slides/posters go in Stage 8.
| Project | Description | Type | Stars | Link | Demo | Paper |
|---|---|---|---|---|---|---|
| PaperBanana | Multi-agent framework for automated academic illustration, generating publication-ready charts from text descriptions | agent | ⭐ 6.6k | GitHub | Demo | arXiv 2025 |
| codex-paper-figure-skill | Codex skill that turns paper sections, method descriptions, and figure concepts into editable draw.io academic figures | skill | ⭐ updating | GitHub | - | - |
Drafting, polishing, citation verification, LaTeX assistance, rebuttal, reviewing. Survey generation goes in Stage 2.
| Project | Description | Type | Stars | Link | Demo | Paper |
|---|---|---|---|---|---|---|
| Academic Research Skills | Claude Code skill suite covering academic writing, polishing, submission checks, and publication workflow | skill | ⭐ 32.5k | GitHub | - | - |
| RefChecker | Academic reference validation tool for checking citation existence, metadata errors, and likely fabricated references | tool | ⭐ 402 | GitHub | - | - |
| Research Paper Lifecycle Skills | Agent Skills package for literature review, citation verification, submission checks, rebuttals, artifacts, slides, and posters | skill | ⭐ updating | GitHub | Website | - |
Code reproduction, demo experience, model & dataset publishing.
| Project | Description | Type | Link | Guide |
|---|---|---|---|---|
| ModelScope Studios | Host research demo reproduction, release, and hands-on experience | platform | ModelScope Studio | Docs |
| ModelScope Notebook | Free cloud Jupyter with GPU, persistent storage, and AI coding assistance | platform | Gallery | Guide |
| ModelScope AI for Science | Open-source AI4Science model hub covering life, earth, material, and social sciences | platform | Nexa Models | Docs |
| Paper2Code | Multi-agent system converting academic papers into runnable code repositories | agent | GitHub | Quick Start |
| data-to-paper | Multi-AI-agent system that autonomously produces verifiable papers from raw data | python pkg | GitHub | - |
| Manage AI Research Projects | Agent Skill for Claude Code/Codex to scaffold reproducible research projects, audit metadata and result traceability, and record AI workflow assets | skill | GitHub | README |
Slides, posters, blog posts, social media outreach, citation analysis, academic influence tools.
| Project | Description | Type | Stars | Link | Demo/Practice | Paper |
|---|---|---|---|---|---|---|
| CitationClaw | Agent-powered explainable citation impact mining for post-publication analysis | tool | ⭐ 307 | GitHub | ModelScope Studio | - |
| ModelScope AI Super Resume | Build your research profile to showcase models, datasets, papers, demos, and academic influence | platform | - | ModelScope | Examples: VoyagerX · chenxie95 | - |
| Title | Summary | Article | Demo |
|---|---|---|---|
| DIY Your Protein — AlphaFold3 Inference | Hands-on tutorial for protein structure prediction with AlphaFold3 on ModelScope | Learn | Studio |
| AI-Ready Remote Sensing: From Open Data to Open-Source Ecosystem | AI-Ready data intro and how ModelScope improves AI readiness for remote sensing research | Learn | - |
| Open Source GeoAI Practice with ModelScope | APGARSS tutorial collection: open-source GeoAI practice based on ModelScope | Learn | - |
| PaperSeek - Literature Search with Natural Language | A practical workflow for natural-language literature search, automatic query expansion, and candidate paper collection | Learn | Studio |
What we look for:
- ✅ Clearly targets scientific research scenarios (not general-purpose AI assistants)
- ✅ Has runnable code or a reusable workflow
- ✅ Actively maintained (updated within the last 6 months)
- ✅ Clear documentation or README for onboarding
- ✅ Addresses real research pain points with demonstrated use cases
- ✅ Open source or has a free tier
Bonus points: backed by a paper · has a live demo · cited by the research community · validated in real projects
Beyond recommending existing open-source projects, we especially encourage you to distill AI-assisted workflows you've actually run in your own research.
Contribution types:
- 🔗 Recommend a Resource: Suggest a tool/paper/agent/skill you've used — just add a row to the relevant stage table with a one-line description of your use case.
- 📝 Share a Best Practice: Share a reusable workflow you've built, for example:
- A real research scenario ("I used it for literature review in XXX, saving ~X hours")
- Failure cases & boundaries ("Doesn't work well in Y scenario because of Z")
How to contribute: Add a row at the end of the relevant stage table, and fill in a link in the "Demo/Practice" column. We recommend publishing on ModelScope Learn with the #vibe-research tag for discoverability; external links are also welcome. If your practice spans multiple stages, place it in the most core stage and note "also covers stage X" in the description.
About the Stars column:
- Stars are automatically maintained by GitHub Action — just add the marker on first submission, and they will be updated automatically
- When adding a new project, fill the
Starscolumn with:For example:`<!--stars:OWNER/REPO-->⭐ updating<!--/stars-->`| DeepScientist | Local-first autonomous research studio... | agent | <!--stars:ResearAI/DeepScientist-->⭐ 3.1k<!--/stars--> | [GitHub](...) | [Demo](...) | [arXiv 2025](...) | - Update frequency:
- Scheduled: Daily at 2:00 AM (UTC)
- Immediate: Triggered right after a PR is merged to
mainifREADME.mdis changed
- If the GitHub API is temporarily unavailable, the script automatically keeps the previous star count — no blank values
Thanks to all contributors for their participation and sharing!
