gudehhh666/Awesome_Scientific_Agent

Awesome Scientific Agent

Python

28 commits

updated Aug 25, 2026

See the code

README

Awesome Scientific Agent

A curated map of LLM-based agents for automated scientific research.

Overview • Taxonomy • Methods • Agentic Repos • Construction • Enhancement • Benchmarks • Contributing • Citation

✨ Overview

Large language model agents are becoming a practical interface for AI for Science (AI4S): reading literature, generating hypotheses, designing experiments, operating tools, analyzing results, and reviewing research outputs. This repository organizes the scientific-agent landscape around the research workflow and the autonomy level of each system.

What this repository maps	How to read it
Agent levels	Assistant, Partner, and Avatar describe increasing autonomy and responsibility in scientific workflows.
Research stages	Literature, Hypothesis, Design, Verification, Analysis, and Evaluation show where each method contributes.
Agent components	Reasoning, memory, and collaboration enhancements highlight how systems are built beyond a base LLM.
Evaluation resources	Benchmarks and datasets help compare scientific-agent capabilities across domains.

Scientific agents across the research lifecycle

📖 Reading Map

Section	Use it for
💡 Taxonomy	Compare representative scientific agents by level, domain, backbone, capability, components, and research stage.
🧩 Method Guides	Read focused method notes for construction, enhancement, evaluation, and auto-research systems.
✈️ Scientific Agents Construction	Find papers and systems about building agent workflows, prompts, tools, context, and domain interfaces.
🚀 Scientific Agents Enhancement	Explore reasoning, memory, workflow, and self-improvement techniques for stronger agents.
⚖️ Benchmark For Scientific Agents	Locate benchmarks for scientific reasoning, code generation, data analysis, citation, and domain evaluation.
🧭 Broader Agentic & Auto-Research Repositories	Track influential repositories beyond scientific agents, including Karpathy-style training stacks, coding agents, browser agents, and orchestration frameworks.

🖼️ Survey Figure Index

The figures below are synchronized from the TPAMI survey materials so the repository mirrors the paper's structure rather than acting only as a paper list.

Figure	What it explains	Repository file
Research lifecycle	How scientific agents support literature, hypothesis, design, verification, analysis, and evaluation.	overall_short.png
Survey organization	High-level organization of the survey and repository map.	overall.png
E/M role taxonomy	Capability envelope and capability maturity, inducing Assistant, Partner, and Avatar roles.	level.png
Extended role view	Additional role-level view used by the survey materials.	level_2.png
Construction overview	Agent construction methodology.	construction_overview.png
Knowledge organization	How scientific agents organize domain knowledge.	knowledge_organization.png
Orchestration	Coordination and workflow orchestration in scientific-agent construction.	orchestration_flat.png
Enhancement overview	Overview of scientific-agent capability enhancement.	enhancement_overview.png
Memory systems	Memory structures for scientific agents.	memory.png
Reasoning enhancement	Reasoning enhancement patterns for scientific agents.	reasoning.png
Benchmark overview	Scientific-agent benchmark and evaluation metric landscape.	benchmark_overview.png

💡 Taxonomy

overall

Taxonomy Dashboard

Level	Role in scientific work	Count	Typical scope
Assistant	Helps with bounded scientific tasks under direct human steering.	33	Literature synthesis, QA, design assistance, and analysis support.
Partner	Collaborates across multiple workflow steps with stronger tool use or feedback loops.	20	Ideation, experiment planning, automation, review, and domain reasoning.
Avatar	Acts as a higher-autonomy research executor in digital or physical environments.	15	Autonomous labs, discovery loops, and end-to-end research.

Capability / component	Meaning
E	Capability envelope: the breadth of scientific workflow coverage.
M	Capability maturity: the maturity of autonomy and execution.
R / Mem. / C	Reasoning enhancement, memory enhancement, and collaboration enhancement.
Stages	Literature, Hypothesis, Design, Verification, Analysis, and Evaluation.

Full taxonomy table — 68 scientific agents across levels, domains, capabilities, components, and research stages

Capability envelope is abbreviated as E, capability maturity as M, reasoning enhancement as R, memory enhancement as Mem., and collaboration enhancement as C.

Level	Method	Domain	LLM Backbone	E	M	R	Mem.	C	Application Stages	Task Description
Assistant	LitLLM	General	General-purpose	E1	M1	No	No	No	Literature, Analysis	Literature review synthesis
Assistant	otto-SR	Medical	General-purpose	E1	M1	Yes	No	No	Literature, Analysis	Systematic review synthesis
Assistant	SciMON	General	General-purpose	E1	M1	Yes	No	No	Literature, Hypothesis	Novel hypothesis generation
Assistant	KG-FM	Materials	General-purpose	E1	M1	Yes	No	No	Literature, Hypothesis, Analysis	Knowledge-grounded materials QA
Assistant	HypoGen	General	General-purpose	E1	M1	Yes	No	No	Hypothesis	Research hypothesis generation
Assistant	LLM-SR	Physics	General-purpose	E1	M1	Yes	Yes	No	Hypothesis, Analysis	Symbolic equation discovery
Assistant	InstructMol	Chemistry	Domain-specialized	E1	M1	No	No	No	Verification, Analysis	Molecular instruction following
Assistant	GeneGPT	Medical	General-purpose	E1	M1	Yes	No	No	Analysis	Genomic question answering
Assistant	TAIS	Medical	General-purpose	E1	M1	Yes	No	Yes	Verification, Analysis	Gene expression analysis
Assistant	DrugAgent	Medical	General-purpose	E1	M1	Yes	No	Yes	Analysis	ML-driven drug discovery
Assistant	DrugGen	Medical	General-purpose	E1	M1	No	No	No	Design, Verification	Targeted molecular generation
Assistant	ChemAgent	Chemistry	General-purpose	E1	M1	Yes	Yes	No	Design, Verification, Analysis	Multi-step chemical reasoning
Assistant	ChatChemTS	Chemistry	Domain-specialized	E1	M1	No	No	No	Design, Verification	Conversational molecule generation
Assistant	PaperQA	General	General-purpose	E1	M2	Yes	Yes	No	Literature, Analysis	Scientific document QA
Assistant	ChatCite	General	General-purpose	E1	M2	Yes	Yes	No	Literature, Analysis	Evidence-aware literature synthesis
Assistant	CoI-Agent	General	General-purpose	E1	M2	Yes	Yes	No	Literature, Hypothesis	Chain-structured ideation
Assistant	Deep Ideation	General	General-purpose	E1	M2	Yes	Yes	No	Hypothesis, Analysis	Concept-network ideation
Assistant	IRIS	General	General-purpose	E1	M2	Yes	No	Yes	Hypothesis, Analysis	Interactive hypothesis search
Assistant	LlaSMol	Chemistry	Domain-specialized	E1	M2	Yes	No	No	Verification	Molecular design assistance
Assistant	Ether0	Chemistry	Domain-specialized	E1	M2	Yes	No	No	Design, Verification	Complex molecular design
Assistant	ChemCrow	Chemistry	General-purpose	E1	M2	Yes	No	No	Literature, Design, Verification	Tool-augmented chemistry assistance
Assistant	HoneyComb	Materials	General-purpose	E1	M2	Yes	No	Yes	Literature, Design, Verification	Materials design assistance
Assistant	PaperCoder	Computer Science	General-purpose	E1	M2	Yes	No	Yes	Literature, Design, Verification	Paper-to-code generation
Assistant	BioResearcher	Biomedical	General-purpose	E2	M1	Yes	No	No	Literature, Hypothesis, Design, Verification, Analysis	Biological workflow assistance
Assistant	ProtAgents	Biology	General-purpose	E2	M1	Yes	No	Yes	Hypothesis, Design, Verification	Protein design loop
Assistant	MOOSE-Chem	Chemistry	General-purpose	E2	M1	No	Yes	No	Hypothesis, Verification	Chemical hypothesis generation
Assistant	Meta-OpenFoam	Physics	General-purpose	E2	M1	Yes	No	Yes	Design, Verification, Analysis	CFD workflow orchestration
Assistant	FoamAgent	Physics	General-purpose	E2	M1	Yes	No	Yes	Design, Verification, Analysis	Natural-language CFD execution
Assistant	PiFlow	General	General-purpose	E2	M1	Yes	No	Yes	Hypothesis, Design, Verification, Analysis	Principle-guided experiment loops
Assistant	DrBioRight 2.0	Biology	General-purpose	E2	M1	Yes	No	No	Verification, Analysis	Bioinformatics workflow analysis
Assistant	OriGene	Medical	General-purpose	E2	M1	Yes	Yes	Yes	Hypothesis, Verification, Analysis	Target discovery and validation
Assistant	CellVoyager	Biology	General-purpose	E2	M1	Yes	No	No	Hypothesis, Design, Verification, Analysis	Autonomous scRNA-seq analysis
Assistant	VASPilot	Materials	General-purpose	E2	M1	Yes	No	Yes	Design, Verification	Autonomous DFT execution
Partner	DARWIN 1.5	Biology/Chemistry	General-purpose	E2	M2	No	No	No	Literature, Verification, Analysis	Domain reasoning and analysis
Partner	Crispr-GPT	Biology	General-purpose	E2	M2	Yes	Yes	No	Hypothesis, Design, Evaluation	CRISPR design assistance
Partner	Chemma	Chemistry	General-purpose	E2	M2	Yes	No	No	Literature, Hypothesis, Design	Property-guided synthesis planning
Partner	MRAgent	Medical	General-purpose	E2	M2	Yes	No	No	Literature, Design, Verification, Analysis	MR-based medical inference
Partner	Aviary	Hybrid	General-purpose	E2	M2	Yes	No	No	Literature, Hypothesis, Design, Verification, Analysis	General scientific assistance
Partner	Virtual Lab	General	General-purpose	E2	M2	Yes	No	Yes	Literature, Hypothesis, Design, Verification, Analysis, Evaluation	PI-guided virtual experimentation
Partner	DeepRare	Medicine	General-purpose	E2	M2	Yes	Yes	Yes	Literature, Hypothesis, Analysis, Evaluation	Rare-disease differential diagnosis
Partner	MatPilot	Materials	General-purpose	E2	M2	Yes	No	Yes	Literature, Hypothesis, Design, Verification, Analysis	Language-driven materials design
Partner	SciToolAgent	General	General-purpose	E2	M2	Yes	No	Yes	Literature, Hypothesis, Design, Verification, Analysis	Tool-grounded scientific reasoning
Partner	Organa	Chemistry	General-purpose	E2	M2	No	No	No	Design, Verification	Human-guided robotic chemistry
Partner	FunSearch	Mathematics	General-purpose	E2	M2	Yes	Yes	No	Hypothesis, Design, Analysis	Evolutionary mathematical discovery
Partner	StarWhisper	Astronomy	General-purpose	E2	M2	Yes	Yes	Yes	Design, Verification	Autonomous telescope operations
Partner	CycleResearcher	Computer Science	General-purpose	E2	M2	Yes	No	Yes	Literature, Hypothesis, Evaluation	Iterative paper improvement
Partner	Biomni	Biology	General-purpose	E2	M2	Yes	Yes	Yes	Literature, Hypothesis, Design, Verification, Analysis	Broad biological automation
Partner	SciAgents	General	General-purpose	E2	M2	Yes	Yes	Yes	Literature, Hypothesis, Analysis	KG-guided scientific discovery
Partner	AI Scientist	Computer Science	General-purpose	E3	M1	No	No	No	Hypothesis, Design, Verification, Analysis, Evaluation	End-to-end CS research
Partner	AI-Researcher	Computer Science	General-purpose	E3	M1	Yes	No	Yes	Literature, Hypothesis, Analysis, Evaluation	End-to-end research assistance
Partner	Agentrxiv	Computer Science	General-purpose	E3	M1	Yes	Yes	Yes	Literature, Hypothesis, Verification, Analysis	Preprint-grounded agent research
Partner	Agent Laboratory	Computer Science	General-purpose	E3	M1	Yes	Yes	Yes	Literature, Hypothesis, Design, Analysis, Evaluation	End-to-end CS experimentation
Avatar	A-Lab	Materials	General-purpose	E2	M3	No	Yes	No	Literature, Hypothesis, Design, Verification, Analysis	Autonomous materials synthesis
Avatar	AlphaEvolve	General	General-purpose	E2	M3	Yes	Yes	No	Hypothesis, Design, Verification	Evolutionary scientific optimization
Avatar	OpenEvidence	Medicine	Domain-specialized	E2	M3	Yes	No	No	Literature, Analysis, Evaluation	Point-of-care clinical decision
Avatar	AILA	Materials	General-purpose	E2	M3	Yes	No	Yes	Design, Verification, Analysis	Autonomous instrument operation
Avatar	MOSAIC-chemistry	Chemistry	General-purpose	E2	M3	Yes	No	Yes	Design, Verification, Analysis	Collective synthesis planning
Avatar	MARS	Materials	General-purpose	E2	M3	Yes	No	Yes	Literature, Hypothesis, Design, Verification, Analysis	Robotic materials discovery
Avatar	ScienceOne	Biology	Domain-specialized	E3	M2	Yes	Yes	Yes	Literature, Hypothesis, Design, Verification, Analysis, Evaluation	End-to-end scientific automation
Avatar	AI co-scientist	General	General-purpose	E3	M2	Yes	Yes	Yes	Literature, Hypothesis, Design, Verification, Analysis	Multi-agent scientific co-discovery
Avatar	AI Scientist-v2	Computer Science	General-purpose	E3	M2	Yes	No	No	Hypothesis, Design, Verification, Analysis, Evaluation	Workshop-level CS research
Avatar	Coscientist	Chemistry	General-purpose	E3	M2	Yes	No	No	Literature, Hypothesis, Design, Verification, Analysis	Autonomous chemistry execution
Avatar	Robin	Biology	General-purpose	E3	M2	Yes	Yes	Yes	Literature, Hypothesis, Design, Verification, Analysis, Evaluation	Multi-agent biological discovery
Avatar	Sparks	Biology	General-purpose	E3	M2	Yes	Yes	Yes	Hypothesis, Design, Verification, Analysis, Evaluation	Multi-agent protein design
Avatar	InternAgent-1.5	General	Domain-specialized	E3	M2	Yes	Yes	Yes	Literature, Hypothesis, Design, Verification, Analysis	Long-horizon scientific discovery
Partner	SR-Scientist	Physics/Mathematics	General-purpose	E2	M2	Yes	No	No	Hypothesis, Design, Verification, Analysis	Long-horizon agentic scientific equation discovery
Avatar	EvoScientist	Computer Science	General-purpose	E3	M2	Yes	Yes	Yes	Literature, Hypothesis, Design, Verification, Analysis, Evaluation	Self-evolving multi-agent end-to-end discovery
Avatar	Self-Evolving Fluid Control Agent	Physics/Engineering	General-purpose	E2	M3	Yes	No	No	Hypothesis, Design, Verification, Analysis	Autonomous physically reasoned controller discovery

Level 1: Agent As Assistant

Level 1: Agent as Assistant - bounded task support and domain-specific assistance

AstroLLaMA‑Chat: Scaling AstroLLaMA with Conversational and Diverse Datasets — arXiv, 2024 · paper
BioGPT: Generative Pre‑trained Transformer for Biomedical Text Generation and Mining — arXiv, 2022 · paper
DARWIN 1.5: Large Language Models as Materials‑Science Foundation Models — arXiv, 2024 · paper
ChemBERTa: Large‑Scale Self‑Supervised Pre‑training for Molecular Property Prediction — arXiv, 2020 · paper
ChemAU: Harness the Reasoning of LLMs in Chemical Research with Adaptive Uncertainty Estimation — arXiv, 2025 · paper
ChemDFM: A Large Language Foundation Model for Chemistry — arXiv, 2024 · paper
LlaSMol: Advancing Large Language Models for Chemistry with a Large‑Scale, Comprehensive, High‑Quality Instruction Tuning Dataset — arXiv, 2024 · paper
InstructMol: Multi‑modal Integration for Building a Versatile and Reliable Molecular Assistant in Drug Discovery — arXiv, 2023 · paper
ether0: A Scientific Reasoning Model for Chemistry — arXiv, 2025 · paper
Leveraging Large Language Models for Predictive Chemistry — Nature, 2024 · paper
Multi‑modal Molecule Structure–Text Model for Text‑based Retrieval and Editing — Nature, 2023 · paper
ClimateGPT: Towards AI Synthesizing Interdisciplinary Research on Climate Change — arXiv, 2024 · paper
Sparks of Science: Hypothesis Generation Using Structured Paper Data — arXiv, 2025 · paper
DeepSeek‑Prover‑V2: Advancing Formal Mathematical Reasoning via Reinforcement Learning for Subgoal Decomposition — arXiv, 2025 · paper
BiMediX: Bilingual Medical Mixture of Experts LLM — arXiv, 2024 · paper
ChatDoctor: A Medical Chat Model Fine‑tuned on a Large Language Model (LLAMA) Using Medical Domain Knowledge — arXiv, 2023 · paper
AgentMD: Empowering Language Agents for Risk Prediction with Large‑Scale Clinical Tool Learning — arXiv, 2024 · paper
MedAlpaca: An Open‑Source Collection of Medical Conversational AI Models and Training Data — arXiv, 2023 · paper
DrugGen Enhances Drug Discovery with Large Language Models and Reinforcement Learning — Nature, 2025 · paper
LLM‑SR: Scientific Equation Discovery via Programming with Large Language Models — arXiv, 2024 · paper
LitLLM: A Toolkit for Scientific Literature Review — arXiv, 2024 · paper
SciBERT: A Pretrained Language Model for Scientific Text — arXiv, 2019 · paper
SciMON: Scientific Inspiration Machines Optimized for Novelty — arXiv, 2023 · paper
SCITUNE: Aligning Large Language Models with Scientific Multimodal Instructions — arXiv, 2023 · paper
NatureLM: Deciphering the Language of Nature for Scientific Discovery — arXiv, 2025 · paper

Level 2: Agent As Partner

Level 2: Agent as Partner - multi-step collaboration and workflow orchestration

StarWhisper Telescope: Agent‑Based Observation Assistant System to Approach an AI Astrophysicist — arXiv, 2024 · paper
From Intention to Implementation: Automating Biomedical Research via LLMs — arXiv, 2024 · paper
CRISPR‑GPT: An LLM Agent for Automated Design of Gene‑Editing Experiments — arXiv, 2024 · paper
Towards an AI Co‑Scientist — arXiv, 2025 · paper
DrBioRight 2.0: An LLM‑Powered Bioinformatics Chatbot for Large‑Scale Cancer Functional Proteomics Analysis — Nature, 2025 · paper
ProtAgents: Protein Discovery via Large Language Model Multi‑Agent Collaboration — arXiv, 2024 · paper
MOOSE‑Chem: Large Language Models for Rediscovering Unseen Chemistry Scientific Hypotheses — arXiv, 2024 · paper
Autonomous Chemical Research with Large Language Models — Nature, 2023 · paper
ChemCrow: Augmenting Large‑Language Models with Chemistry Tools — arXiv, 2023 · paper
ORGANA: A Robotic Assistant for Automated Chemistry Experimentation and Characterization — arXiv, 2024 · paper
xChemAgents: Agentic AI for Explainable Quantum Chemistry — arXiv, 2025 · paper
ChemAgent: Self‑Updating Memories in Large Language Models Improves Chemical Reasoning — arXiv, 2025 · paper
Large Language Models to Accelerate Organic Chemistry Synthesis — arXiv, 2025 · paper
MyCrunchGPT: A ChatGPT‑Assisted Framework for Scientific Machine Learning — arXiv, 2023 · paper
MetaOpenFoam: An LLM‑Based Multi‑Agent Framework for CFD — arXiv, 2024 · paper
FoamAgent: Towards Automated Intelligent CFD Workflows — arXiv, 2025 · paper
AI‑Researcher: Autonomous Scientific Innovation — arXiv, 2025 · paper
Jupybara: Operationalizing a Design Space for Actionable Data Analysis and Storytelling with LLMs — arXiv, 2025 · paper
FlowAgent: Achieving Compliance and Flexibility for Workflow Agents — arXiv, 2025 · paper
The AI Scientist: Towards Fully Automated Open‑Ended Scientific Discovery — arXiv, 2024 · paper
GeoGPT: Understanding and Processing Geospatial Tasks through an Autonomous GPT — arXiv, 2023 · paper
PaperQA: Retrieval‑Augmented Generative Agent for Scientific Research — arXiv, 2023 · paper
ChatCite: LLM Agent with Human Workflow Guidance for Comparative Literature Summary — arXiv, 2024 · paper
PiFlow: Principle‑Aware Scientific Discovery with Multi‑Agent Collaboration — arXiv, 2025 · paper
Aviary: Training Language Agents on Challenging Scientific Tasks — arXiv, 2024 · paper
An Autonomous Laboratory for the Accelerated Synthesis of Novel Materials — Nature, 2023 · paper
Construction of a Knowledge Graph for Framework Material Enabled by Large Language Models and Its Application — npjCM, 2025 · paper
MultiCrossModal Automated Agent for Integrating Diverse Materials Science Data — arXiv, 2025 · paper
LLMatDesign: Autonomous Materials Discovery with Large Language Models — arXiv, 2024 · paper
Honeycomb: A Flexible LLM‑Based Agent System for Materials Science — arXiv, 2024 · paper
DrugAgent: Automating AI‑Aided Drug Discovery Programming through LLM Multi‑Agent Collaboration — arXiv, 2024 · paper
Toward a Team of AI‑Made Scientists for Scientific Discovery from Gene Expression Data — arXiv, 2024 · paper
MedAgents: Large Language Models as Collaborators for Zero‑Shot Medical Reasoning — arXiv, 2023 · paper
Automation of Systematic Reviews with Large Language Models — medRxiv, 2025 · paper
MRAgent: An LLM‑Based Automated Agent for Causal Knowledge Discovery in Disease via Mendelian Randomization — BriefBioinf, 2025 · website
GeneGPT: Augmenting Large Language Models with Domain Tools for Improved Access to Biomedical Information — arXiv, 2023 · paper
SR-Scientist: Scientific Equation Discovery With Agentic AI — ICLR, 2026 · paper · code
Fantastic Scientific Agents and How to Build Them: AgentBuild for Rietveld Refinement — arXiv, 2026 · paper
SciOrch: Learning to Orchestrate Expert LLMs for Solving Frontier Multimodal Scientific Reasoning Tasks — arXiv, 2026 · paper

Level 3: Agent As Avatar

Level 3: Agent as Avatar - autonomous research execution and long-horizon discovery

CycleResearcher: Improving Automated Research via Automated Review — arXiv, 2024 · paper
The AI Scientist‑v2: Workshop‑Level Automated Scientific Discovery via Agentic Tree Search — arXiv, 2025 · paper
AgentRxiv: Towards Collaborative Autonomous Research — arXiv, 2025 · paper
Agent Laboratory: Using LLM Agents as Research Assistants — arXiv, 2025 · paper
BiOMNI: A General‑Purpose Biomedical AI Agent — bioRxiv, 2025 · paper
OriGene: A Self‑Evolving Virtual Disease Biologist Automating Therapeutic Target Discovery — bioRxiv, 2025 · paper
CellVoyager: AI CompBio Agent Generates New Insights by Autonomously Analyzing Biological Data — bioRxiv, 2025 · paper
Sparks: Multi‑Agent Artificial Intelligence Model Discovers Protein Design Principles — arXiv, 2025 · paper
AlphaEvolve: A Coding Agent for Scientific and Algorithmic Discovery — arXiv, 2025 · paper
Robin: A Multi‑Agent System for Automating Scientific Discovery — arXiv, 2025 · paper
ScienceOne — Project, 2025 · website
EvoScientist: Towards Multi-Agent Evolving AI Scientists for End-to-End Scientific Discovery — arXiv, 2026 · paper · code
Self-Evolving Scientific Agent Discovers Generalizable Physically-Reasoned Fluid Control — arXiv, 2026 · paper

🧩 Method Guides

The root README keeps the curated map compact; the method guides provide deeper explanations for readers who want to understand how scientific agents are built, improved, evaluated, and connected to broader auto-research systems.

Guide	What it covers
Construction Methods	Knowledge organization, knowledge injection, tool integration, orchestration, and domain interfaces.
Enhancement Methods	Reasoning, memory, collaboration, workflow search, and self-review.
Evaluation Methods	Benchmark selection, executable evaluation, citation grounding, and long-horizon validation.
Auto-Research Systems	End-to-end research loops, coding/browser agents, training resources, and orchestration frameworks.

🧭 Broader Agentic & Auto-Research Repositories

Beyond scientific-agent papers, the broader agent ecosystem is moving quickly across model training, software engineering, web automation, and multi-agent orchestration. This watchlist keeps a lightweight bridge from the survey taxonomy to practical repositories that shape how autonomous research and agentic systems are built.

Repository	Scope	Why follow it
karpathy/nanochat	Minimal end-to-end LLM training and chat stack	Tracks Karpathy's compact, hackable path from tokenizer and pretraining to finetuning, evaluation, inference, and chat UI.
karpathy/llm.c	LLM training in C/CUDA	Useful for understanding low-level training kernels, performance constraints, and reproducible GPT-style training.
SakanaAI/AI-Scientist	Automated idea-to-paper research loop	A reference point for autonomous ideation, coding, experimentation, paper writing, and automated review.
SakanaAI/AI-Scientist-v2	Agentic tree search for automated discovery	Follows the next iteration of AI Scientist with broader exploration and stronger end-to-end workflow design.
SamuelSchmidgall/AgentLaboratory	Human-guided autonomous research assistant	Shows how literature review, experimentation, and report writing can be composed into a full research workflow.
NoviScl/AI-Researcher	Research ideation and execution studies	Provides agent pipelines and human-study data for comparing LLM-generated ideas with expert research ideas.
OpenHands/OpenHands	AI software development agents	A generalist coding-agent platform for editing repositories, using terminals, browsing, and operating in sandboxed environments.
SWE-agent/SWE-agent	GitHub issue fixing and SWE-bench agents	A practical baseline for agentic software engineering, debugging, and repository-level task execution.
huggingface/smolagents	Lightweight code-agent framework	Good for studying minimal abstractions, code-as-action agents, sandboxed execution, and open-model agent workflows.
microsoft/autogen	Agentic AI programming framework	Useful for multi-agent conversations, orchestration patterns, and prototyping collaborative agent systems.
crewAIInc/crewAI	Multi-agent orchestration	Focuses on role-based agents, task delegation, crews, flows, and production-style automation workflows.
langchain-ai/langgraph	Graph-based agent workflows	Useful for durable, stateful, controllable agent graphs and long-running workflow orchestration.
FoundationAgents/MetaGPT	Multi-agent software company metaphor	A representative multi-agent framework for decomposing product/software work into role-specialized agents.
browser-use/browser-use	Browser automation for agents	Tracks web interaction patterns, browser control, and task automation over ordinary websites.
Significant-Gravitas/AutoGPT	Early autonomous agent platform	Still useful as historical context for goal-directed agents, autonomous task decomposition, and agent productization.
EvoScientist/EvoScientist	Self-evolving end-to-end AI scientist	Tracks persistent research memory, multi-agent experimentation, and human-on-the-loop research workflows.

✈️ Scientific Agents Construction

Agent construction methodology

Knowledge organization in scientific agents Orchestration and coordination in scientific-agent construction

Construction resources - prompts, context, tools, workflows, and domain interfaces

PromptAgent: Strategic Planning with Language Models Enables Expert-level Prompt Optimization — arXiv, 2023 · paper
Context Engineering: A Practical Handbook for Context Design, Orchestration, and Optimization — GitHub, 2025 · code
Verl: Volcano Engine Reinforcement Learning for Large Language Models — GitHub, 2024 · code
GraphRAG: A Modular Graph-Based Retrieval-Augmented Generation (RAG) System — arXiv, 2023 · paper
ChemCrow: Augmenting Large‑Language Models with Chemistry Tools — arXiv, 2023 · paper
Virtual Lab: AI Agents Design New SARS-CoV-2 Nanobodies with Experimental Validation — arXiv, 2024 · paper
OctoTools: An Agentic Framework with Extensible Tools for Complex Reasoning — arXiv, 2025 · paper
FoamAgent: Towards Automated Intelligent CFD Workflows — arXiv, 2025 · paper
MetaOpenFoam: An LLM-Based Multi-Agent Framework for CFD — arXiv, 2024 · paper
AutoPrompt: Eliciting Knowledge from Language Models with Automatically Generated Prompts — EMNLP, 2020 · paper
InstructBio: Instruction Tuning for Biomedical LLMs — arXiv, 2024 · paper
LangChain: Building Applications with LLMs through Composability — GitHub, 2022 · code
LightRAG: Simple and Fast Retrieval‑Augmented Generation — arXiv, 2024 · paper
SciTUNE: Aligning Large Language Models with Scientific Multimodal Instructions — arXiv, 2023 · paper
ClimateGPT: Towards AI Synthesizing Interdisciplinary Research on Climate Change — arXiv, 2024 · paper
SciMON: Scientific Inspiration Machines Optimized for Novelty — arXiv, 2023 · paper
TAIS: Gene Expression Agent with LLMs — arXiv, 2025 · paper
StarWhisper Telescope: Agent‑Based Observation Assistant System to Approach an AI Astrophysicist — arXiv, 2024 · paper
Fantastic Scientific Agents and How to Build Them: AgentBuild for Rietveld Refinement — arXiv, 2026 · paper

🚀 Scientific Agents Enhancement

Scientific-agent ability enhancement

Scientific agents' memory systems Illustration of scientific agent reasoning enhancement

Enhancement resources - reasoning, memory, workflow search, reflection, and collaboration

MemGPT: Toward LLMs as Operating Systems — arXiv, 2023 · paper
AFlow: Automating Agentic Workflow Generation — arXiv, 2023 · paper
ChemAgent: Self‑Updating Memories in Large Language Models Improves Chemical Reasoning — ICLR, 2025 · paper
DeepSeek‑Prover‑V2: Advancing Formal Mathematical Reasoning — arXiv, 2025 · paper
LLM‑SR: Scientific Equation Discovery via Programming with LLMs — ICLR, 2025 · paper
Sparks: Multi‑Agent Artificial Intelligence Model Discovers Protein Design Principles — arXiv, 2025 · paper
ether0: A Scientific Reasoning Model for Chemistry — arXiv, 2025 · paper
Robin: A Multi‑Agent System for Automating Scientific Discovery — arXiv, 2025 · paper
AgentRxiv: Towards Collaborative Autonomous Research — arXiv, 2025 · paper
ChatGPT Research Group for Optimizing the Crystallinity of MOFs and COFs — ACS, 2023 · paper
AI Co-Scientist: Towards an AI Co-Scientist — arXiv, 2025 · paper
ReAct: Synergizing Reasoning and Acting in Language Models — arXiv, 2022 · paper
Reflexion: Language Agents with Verbal Reinforcement Learning — NeurIPS, 2023 · paper
Self‑Refine: Iterative Self‑Improvement with Self‑Feedback — arXiv, 2023 · paper
Self‑Consistency: Reliable Decoding for Complex Reasoning — ICLR, 2023 · paper
AquilaChat: Agent with Long-Context Scratchpad Memory — arXiv, 2024 · paper
EvoScientist: Towards Multi-Agent Evolving AI Scientists for End-to-End Scientific Discovery — arXiv, 2026 · paper
SciOrch: Learning to Orchestrate Expert LLMs for Solving Frontier Multimodal Scientific Reasoning Tasks — arXiv, 2026 · paper
Self-Evolving Scientific Agent Discovers Generalizable Physically-Reasoned Fluid Control — arXiv, 2026 · paper

⚖️ Benchmark For Scientific Agents

Use this section as an evaluation map rather than a flat benchmark list. The resources below cover different failure modes of scientific agents: domain knowledge, executable experiments, citation grounding, data analysis, and long-horizon discovery.

Scientific-agent benchmark and evaluation landscape

Evaluation angle	Representative focus	Useful when you need to test...
Scientific knowledge and reasoning	BioMaze, SuperGPQA, Humanity's Last Exam, MR-Ben	Whether an agent can reason over expert-level scientific concepts.
Citation and literature grounding	CiteBench, ALCE, SurveyForge	Whether outputs are traceable, evidence-aware, and literature-faithful.
Code, data, and experiment execution	MLAgentBench, DSBench, SciCode, PaperBench	Whether an agent can implement, run, debug, and reproduce research workflows.
Domain and embodied environments	DiscoveryWorld, AgentClinic, GenoTEX, LLM-SRBench	Whether an agent performs in domain-specific or simulated scientific settings.

Benchmark resources — evaluation suites for scientific reasoning, data analysis, citation, coding, and agentic discovery

BioMaze: Benchmarking and Enhancing Large Language Models for Biological Pathway Reasoning — arXiv, 2025 · paper · dataset
BioKGBench: A Knowledge Graph Checking Benchmark of AI Agent for Biomedical Science — GitHub, 2025 · code
SurveyForge: On the Outline Heuristics, Memory-Driven Generation, and Multi-Dimensional Evaluation for Automated Survey Writing — arXiv, 2025 · paper · code
MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmark — NeurIPS, 2024 · paper · dataset
Humanity's Last Exam: A Hard Benchmark at the Frontier of Human Knowledge — arXiv, 2025 · paper · website
SuperGPQA: Scaling LLM Evaluation Across 285 Graduate Disciplines — arXiv, 2025 · paper · dataset
CiteBench: A Benchmark for Scientific Citation Text Generation — arXiv, 2022 · paper · code
ALCE: Enabling Large Language Models to Generate Text With Citations — arXiv, 2023 · paper · code
Tomato-Chem: Large Language Models for Rediscovering Unseen Chemistry Scientific Hypotheses — arXiv, 2024 · paper · code
Reviewer2: Optimizing Review Generation Through Prompt Generation — arXiv, 2024 · paper · code
Scientists' First Exam: Probing Cognitive Abilities of MLLM via Perception, Understanding, and Reasoning — arXiv, 2025 · dataset · paper
MR-Ben: A Comprehensive Meta-Reasoning Benchmark for Large Language Models — arXiv, 2024 · paper · website
FigureQA: An Annotated Figure Dataset for Visual Reasoning — arXiv, 2017 · paper · dataset
SciEval: A Multi-Level Large Language Model Evaluation Benchmark for Scientific Research — AAAI, 2024 · paper · dataset
MedXpertQA: Benchmarking Expert-Level Medical Reasoning and Understanding — arXiv, 2025 · paper · dataset
LLM-SRBench: A New Benchmark for Scientific Equation Discovery With Large Language Models — arXiv, 2025 · paper · dataset
GenoTEX: A Benchmark for Evaluating LLM-Based Exploration of Gene Expression Data in Alignment With Bioinformaticians — arXiv, 2024 · paper · dataset
MLAgentBench: Evaluating Language Agents on Machine Learning Experimentation — arXiv, 2023 · paper · code
PaperBench: Evaluating AI's Ability to Replicate AI Research — arXiv, 2025 · paper
DSCodeBench: A Realistic Benchmark for Data Science Code Generation — AAAI, 2026 · paper · code
DiscoveryWorld: A Virtual Environment for Developing and Evaluating Automated Scientific Discovery Agents — arXiv, 2024 · paper · code
SciCode: A Scientist-Curated Benchmark for Scientific Code Generation — arXiv, 2024 · paper · code
AgentClinic: A Multimodal Agent Benchmark to Evaluate AI in Simulated Clinical Environments — arXiv, 2024 · code · paper
SDRBench: Scientific Data Reduction Benchmark for Lossy Compressors — Website, 2021 · website · website
DSBench: How Far Are Data Science Agents from Becoming Data Science Experts? — GitHub, 2025 · code · paper
AISB: AI Scientist Benchmark — NLPCC, 2026 · evaluation kit
DSAgentBench: Can Agents Automate End-to-End Data-Science Workflows in Real Computer Environments? — arXiv, 2026 · paper · code

🤝 Contributing

Catalog tables and lists are generated from structured files in data/. See CONTRIBUTING.md for the evidence and maintenance workflow, and the catalog schema for field definitions. Edit templates/README.md for narrative content, then regenerate this file with python3 scripts/render_readme.py.

🌞 Citation

@article{wang2025hitchhiker,
  title={The Hitchhiker's Guide to Autonomous Research: A Survey of Scientific Agents},
  author={Wang, Xinming and Xu, Jian and Feng, Aslan H and Chen, Yi and Guo, Haiyang and Zhu, Fei and Shao, Yuanqi and Ren, Minsi and Yi, Hongzhu and Lian, Sheng and others},
  year={2025}
}

Contributors

gudehhh666

27 commits

banjiuyufen

1 commits

Similar repos

OSU-NLP-Group/awesome-agents4science

A curated list of papers on LLMs and agents for scientific research and development

zoedsy/awesome-science-agents

AI agents for scientific discovery: papers, tools, and an interactive bilingual atlas with detailed…

du-nlp-lab/LLM4SR

LLM for Scientific Research Survey

133

ResearAI/Awesome-AI-Scientist

This is a survey of research on AI scientists, AI researchers, AI engineers, and a series of…

314

Omni-Scientist/Awesome-AI-Scientist

🧪 Awesome list of AI Scientist papers, systems, benchmarks, datasets and open-source platforms.

103

worldbench/awesome-ai-auto-research

🔥 A Survey on AI Auto-Research

533

ur-whitelab/LLMs-in-science

298

WuizaKaseiyo/awesome-deep-research-agent

638

Languages

Python

100.0%

gudehhh666/Awesome_Scientific_Agent

Awesome Scientific Agent

Python

28 commits

updated Aug 25, 2026

See the code

README

Awesome Scientific Agent

A curated map of LLM-based agents for automated scientific research.

Overview • Taxonomy • Methods • Agentic Repos • Construction • Enhancement • Benchmarks • Contributing • Citation

✨ Overview

What this repository maps	How to read it
Agent levels	Assistant, Partner, and Avatar describe increasing autonomy and responsibility in scientific workflows.
Research stages	Literature, Hypothesis, Design, Verification, Analysis, and Evaluation show where each method contributes.
Agent components	Reasoning, memory, and collaboration enhancements highlight how systems are built beyond a base LLM.
Evaluation resources	Benchmarks and datasets help compare scientific-agent capabilities across domains.

Scientific agents across the research lifecycle

📖 Reading Map

Section	Use it for
💡 Taxonomy	Compare representative scientific agents by level, domain, backbone, capability, components, and research stage.
🧩 Method Guides	Read focused method notes for construction, enhancement, evaluation, and auto-research systems.
✈️ Scientific Agents Construction	Find papers and systems about building agent workflows, prompts, tools, context, and domain interfaces.
🚀 Scientific Agents Enhancement	Explore reasoning, memory, workflow, and self-improvement techniques for stronger agents.
⚖️ Benchmark For Scientific Agents	Locate benchmarks for scientific reasoning, code generation, data analysis, citation, and domain evaluation.
🧭 Broader Agentic & Auto-Research Repositories	Track influential repositories beyond scientific agents, including Karpathy-style training stacks, coding agents, browser agents, and orchestration frameworks.

🖼️ Survey Figure Index

The figures below are synchronized from the TPAMI survey materials so the repository mirrors the paper's structure rather than acting only as a paper list.

Figure	What it explains	Repository file
Research lifecycle	How scientific agents support literature, hypothesis, design, verification, analysis, and evaluation.	overall_short.png
Survey organization	High-level organization of the survey and repository map.	overall.png
E/M role taxonomy	Capability envelope and capability maturity, inducing Assistant, Partner, and Avatar roles.	level.png
Extended role view	Additional role-level view used by the survey materials.	level_2.png
Construction overview	Agent construction methodology.	construction_overview.png
Knowledge organization	How scientific agents organize domain knowledge.	knowledge_organization.png
Orchestration	Coordination and workflow orchestration in scientific-agent construction.	orchestration_flat.png
Enhancement overview	Overview of scientific-agent capability enhancement.	enhancement_overview.png
Memory systems	Memory structures for scientific agents.	memory.png
Reasoning enhancement	Reasoning enhancement patterns for scientific agents.	reasoning.png
Benchmark overview	Scientific-agent benchmark and evaluation metric landscape.	benchmark_overview.png

💡 Taxonomy

overall

Taxonomy Dashboard

Level	Role in scientific work	Count	Typical scope
Assistant	Helps with bounded scientific tasks under direct human steering.	33	Literature synthesis, QA, design assistance, and analysis support.
Partner	Collaborates across multiple workflow steps with stronger tool use or feedback loops.	20	Ideation, experiment planning, automation, review, and domain reasoning.
Avatar	Acts as a higher-autonomy research executor in digital or physical environments.	15	Autonomous labs, discovery loops, and end-to-end research.

Capability / component	Meaning
E	Capability envelope: the breadth of scientific workflow coverage.
M	Capability maturity: the maturity of autonomy and execution.
R / Mem. / C	Reasoning enhancement, memory enhancement, and collaboration enhancement.
Stages	Literature, Hypothesis, Design, Verification, Analysis, and Evaluation.

Full taxonomy table — 68 scientific agents across levels, domains, capabilities, components, and research stages

Capability envelope is abbreviated as E, capability maturity as M, reasoning enhancement as R, memory enhancement as Mem., and collaboration enhancement as C.

Level	Method	Domain	LLM Backbone	E	M	R	Mem.	C	Application Stages	Task Description
Assistant	LitLLM	General	General-purpose	E1	M1	No	No	No	Literature, Analysis	Literature review synthesis
Assistant	otto-SR	Medical	General-purpose	E1	M1	Yes	No	No	Literature, Analysis	Systematic review synthesis
Assistant	SciMON	General	General-purpose	E1	M1	Yes	No	No	Literature, Hypothesis	Novel hypothesis generation
Assistant	KG-FM	Materials	General-purpose	E1	M1	Yes	No	No	Literature, Hypothesis, Analysis	Knowledge-grounded materials QA
Assistant	HypoGen	General	General-purpose	E1	M1	Yes	No	No	Hypothesis	Research hypothesis generation
Assistant	LLM-SR	Physics	General-purpose	E1	M1	Yes	Yes	No	Hypothesis, Analysis	Symbolic equation discovery
Assistant	InstructMol	Chemistry	Domain-specialized	E1	M1	No	No	No	Verification, Analysis	Molecular instruction following
Assistant	GeneGPT	Medical	General-purpose	E1	M1	Yes	No	No	Analysis	Genomic question answering
Assistant	TAIS	Medical	General-purpose	E1	M1	Yes	No	Yes	Verification, Analysis	Gene expression analysis
Assistant	DrugAgent	Medical	General-purpose	E1	M1	Yes	No	Yes	Analysis	ML-driven drug discovery
Assistant	DrugGen	Medical	General-purpose	E1	M1	No	No	No	Design, Verification	Targeted molecular generation
Assistant	ChemAgent	Chemistry	General-purpose	E1	M1	Yes	Yes	No	Design, Verification, Analysis	Multi-step chemical reasoning
Assistant	ChatChemTS	Chemistry	Domain-specialized	E1	M1	No	No	No	Design, Verification	Conversational molecule generation
Assistant	PaperQA	General	General-purpose	E1	M2	Yes	Yes	No	Literature, Analysis	Scientific document QA
Assistant	ChatCite	General	General-purpose	E1	M2	Yes	Yes	No	Literature, Analysis	Evidence-aware literature synthesis
Assistant	CoI-Agent	General	General-purpose	E1	M2	Yes	Yes	No	Literature, Hypothesis	Chain-structured ideation
Assistant	Deep Ideation	General	General-purpose	E1	M2	Yes	Yes	No	Hypothesis, Analysis	Concept-network ideation
Assistant	IRIS	General	General-purpose	E1	M2	Yes	No	Yes	Hypothesis, Analysis	Interactive hypothesis search
Assistant	LlaSMol	Chemistry	Domain-specialized	E1	M2	Yes	No	No	Verification	Molecular design assistance
Assistant	Ether0	Chemistry	Domain-specialized	E1	M2	Yes	No	No	Design, Verification	Complex molecular design
Assistant	ChemCrow	Chemistry	General-purpose	E1	M2	Yes	No	No	Literature, Design, Verification	Tool-augmented chemistry assistance
Assistant	HoneyComb	Materials	General-purpose	E1	M2	Yes	No	Yes	Literature, Design, Verification	Materials design assistance
Assistant	PaperCoder	Computer Science	General-purpose	E1	M2	Yes	No	Yes	Literature, Design, Verification	Paper-to-code generation
Assistant	BioResearcher	Biomedical	General-purpose	E2	M1	Yes	No	No	Literature, Hypothesis, Design, Verification, Analysis	Biological workflow assistance
Assistant	ProtAgents	Biology	General-purpose	E2	M1	Yes	No	Yes	Hypothesis, Design, Verification	Protein design loop
Assistant	MOOSE-Chem	Chemistry	General-purpose	E2	M1	No	Yes	No	Hypothesis, Verification	Chemical hypothesis generation
Assistant	Meta-OpenFoam	Physics	General-purpose	E2	M1	Yes	No	Yes	Design, Verification, Analysis	CFD workflow orchestration
Assistant	FoamAgent	Physics	General-purpose	E2	M1	Yes	No	Yes	Design, Verification, Analysis	Natural-language CFD execution
Assistant	PiFlow	General	General-purpose	E2	M1	Yes	No	Yes	Hypothesis, Design, Verification, Analysis	Principle-guided experiment loops
Assistant	DrBioRight 2.0	Biology	General-purpose	E2	M1	Yes	No	No	Verification, Analysis	Bioinformatics workflow analysis
Assistant	OriGene	Medical	General-purpose	E2	M1	Yes	Yes	Yes	Hypothesis, Verification, Analysis	Target discovery and validation
Assistant	CellVoyager	Biology	General-purpose	E2	M1	Yes	No	No	Hypothesis, Design, Verification, Analysis	Autonomous scRNA-seq analysis
Assistant	VASPilot	Materials	General-purpose	E2	M1	Yes	No	Yes	Design, Verification	Autonomous DFT execution
Partner	DARWIN 1.5	Biology/Chemistry	General-purpose	E2	M2	No	No	No	Literature, Verification, Analysis	Domain reasoning and analysis
Partner	Crispr-GPT	Biology	General-purpose	E2	M2	Yes	Yes	No	Hypothesis, Design, Evaluation	CRISPR design assistance
Partner	Chemma	Chemistry	General-purpose	E2	M2	Yes	No	No	Literature, Hypothesis, Design	Property-guided synthesis planning
Partner	MRAgent	Medical	General-purpose	E2	M2	Yes	No	No	Literature, Design, Verification, Analysis	MR-based medical inference
Partner	Aviary	Hybrid	General-purpose	E2	M2	Yes	No	No	Literature, Hypothesis, Design, Verification, Analysis	General scientific assistance
Partner	Virtual Lab	General	General-purpose	E2	M2	Yes	No	Yes	Literature, Hypothesis, Design, Verification, Analysis, Evaluation	PI-guided virtual experimentation
Partner	DeepRare	Medicine	General-purpose	E2	M2	Yes	Yes	Yes	Literature, Hypothesis, Analysis, Evaluation	Rare-disease differential diagnosis
Partner	MatPilot	Materials	General-purpose	E2	M2	Yes	No	Yes	Literature, Hypothesis, Design, Verification, Analysis	Language-driven materials design
Partner	SciToolAgent	General	General-purpose	E2	M2	Yes	No	Yes	Literature, Hypothesis, Design, Verification, Analysis	Tool-grounded scientific reasoning
Partner	Organa	Chemistry	General-purpose	E2	M2	No	No	No	Design, Verification	Human-guided robotic chemistry
Partner	FunSearch	Mathematics	General-purpose	E2	M2	Yes	Yes	No	Hypothesis, Design, Analysis	Evolutionary mathematical discovery
Partner	StarWhisper	Astronomy	General-purpose	E2	M2	Yes	Yes	Yes	Design, Verification	Autonomous telescope operations
Partner	CycleResearcher	Computer Science	General-purpose	E2	M2	Yes	No	Yes	Literature, Hypothesis, Evaluation	Iterative paper improvement
Partner	Biomni	Biology	General-purpose	E2	M2	Yes	Yes	Yes	Literature, Hypothesis, Design, Verification, Analysis	Broad biological automation
Partner	SciAgents	General	General-purpose	E2	M2	Yes	Yes	Yes	Literature, Hypothesis, Analysis	KG-guided scientific discovery
Partner	AI Scientist	Computer Science	General-purpose	E3	M1	No	No	No	Hypothesis, Design, Verification, Analysis, Evaluation	End-to-end CS research
Partner	AI-Researcher	Computer Science	General-purpose	E3	M1	Yes	No	Yes	Literature, Hypothesis, Analysis, Evaluation	End-to-end research assistance
Partner	Agentrxiv	Computer Science	General-purpose	E3	M1	Yes	Yes	Yes	Literature, Hypothesis, Verification, Analysis	Preprint-grounded agent research
Partner	Agent Laboratory	Computer Science	General-purpose	E3	M1	Yes	Yes	Yes	Literature, Hypothesis, Design, Analysis, Evaluation	End-to-end CS experimentation
Avatar	A-Lab	Materials	General-purpose	E2	M3	No	Yes	No	Literature, Hypothesis, Design, Verification, Analysis	Autonomous materials synthesis
Avatar	AlphaEvolve	General	General-purpose	E2	M3	Yes	Yes	No	Hypothesis, Design, Verification	Evolutionary scientific optimization
Avatar	OpenEvidence	Medicine	Domain-specialized	E2	M3	Yes	No	No	Literature, Analysis, Evaluation	Point-of-care clinical decision
Avatar	AILA	Materials	General-purpose	E2	M3	Yes	No	Yes	Design, Verification, Analysis	Autonomous instrument operation
Avatar	MOSAIC-chemistry	Chemistry	General-purpose	E2	M3	Yes	No	Yes	Design, Verification, Analysis	Collective synthesis planning
Avatar	MARS	Materials	General-purpose	E2	M3	Yes	No	Yes	Literature, Hypothesis, Design, Verification, Analysis	Robotic materials discovery
Avatar	ScienceOne	Biology	Domain-specialized	E3	M2	Yes	Yes	Yes	Literature, Hypothesis, Design, Verification, Analysis, Evaluation	End-to-end scientific automation
Avatar	AI co-scientist	General	General-purpose	E3	M2	Yes	Yes	Yes	Literature, Hypothesis, Design, Verification, Analysis	Multi-agent scientific co-discovery
Avatar	AI Scientist-v2	Computer Science	General-purpose	E3	M2	Yes	No	No	Hypothesis, Design, Verification, Analysis, Evaluation	Workshop-level CS research
Avatar	Coscientist	Chemistry	General-purpose	E3	M2	Yes	No	No	Literature, Hypothesis, Design, Verification, Analysis	Autonomous chemistry execution
Avatar	Robin	Biology	General-purpose	E3	M2	Yes	Yes	Yes	Literature, Hypothesis, Design, Verification, Analysis, Evaluation	Multi-agent biological discovery
Avatar	Sparks	Biology	General-purpose	E3	M2	Yes	Yes	Yes	Hypothesis, Design, Verification, Analysis, Evaluation	Multi-agent protein design
Avatar	InternAgent-1.5	General	Domain-specialized	E3	M2	Yes	Yes	Yes	Literature, Hypothesis, Design, Verification, Analysis	Long-horizon scientific discovery
Partner	SR-Scientist	Physics/Mathematics	General-purpose	E2	M2	Yes	No	No	Hypothesis, Design, Verification, Analysis	Long-horizon agentic scientific equation discovery
Avatar	EvoScientist	Computer Science	General-purpose	E3	M2	Yes	Yes	Yes	Literature, Hypothesis, Design, Verification, Analysis, Evaluation	Self-evolving multi-agent end-to-end discovery
Avatar	Self-Evolving Fluid Control Agent	Physics/Engineering	General-purpose	E2	M3	Yes	No	No	Hypothesis, Design, Verification, Analysis	Autonomous physically reasoned controller discovery

Level 1: Agent As Assistant

Level 1: Agent as Assistant - bounded task support and domain-specific assistance

AstroLLaMA‑Chat: Scaling AstroLLaMA with Conversational and Diverse Datasets — arXiv, 2024 · paper
BioGPT: Generative Pre‑trained Transformer for Biomedical Text Generation and Mining — arXiv, 2022 · paper
DARWIN 1.5: Large Language Models as Materials‑Science Foundation Models — arXiv, 2024 · paper
ChemBERTa: Large‑Scale Self‑Supervised Pre‑training for Molecular Property Prediction — arXiv, 2020 · paper
ChemAU: Harness the Reasoning of LLMs in Chemical Research with Adaptive Uncertainty Estimation — arXiv, 2025 · paper
ChemDFM: A Large Language Foundation Model for Chemistry — arXiv, 2024 · paper
LlaSMol: Advancing Large Language Models for Chemistry with a Large‑Scale, Comprehensive, High‑Quality Instruction Tuning Dataset — arXiv, 2024 · paper
InstructMol: Multi‑modal Integration for Building a Versatile and Reliable Molecular Assistant in Drug Discovery — arXiv, 2023 · paper
ether0: A Scientific Reasoning Model for Chemistry — arXiv, 2025 · paper
Leveraging Large Language Models for Predictive Chemistry — Nature, 2024 · paper
Multi‑modal Molecule Structure–Text Model for Text‑based Retrieval and Editing — Nature, 2023 · paper
ClimateGPT: Towards AI Synthesizing Interdisciplinary Research on Climate Change — arXiv, 2024 · paper
Sparks of Science: Hypothesis Generation Using Structured Paper Data — arXiv, 2025 · paper
DeepSeek‑Prover‑V2: Advancing Formal Mathematical Reasoning via Reinforcement Learning for Subgoal Decomposition — arXiv, 2025 · paper
BiMediX: Bilingual Medical Mixture of Experts LLM — arXiv, 2024 · paper
ChatDoctor: A Medical Chat Model Fine‑tuned on a Large Language Model (LLAMA) Using Medical Domain Knowledge — arXiv, 2023 · paper
AgentMD: Empowering Language Agents for Risk Prediction with Large‑Scale Clinical Tool Learning — arXiv, 2024 · paper
MedAlpaca: An Open‑Source Collection of Medical Conversational AI Models and Training Data — arXiv, 2023 · paper
DrugGen Enhances Drug Discovery with Large Language Models and Reinforcement Learning — Nature, 2025 · paper
LLM‑SR: Scientific Equation Discovery via Programming with Large Language Models — arXiv, 2024 · paper
LitLLM: A Toolkit for Scientific Literature Review — arXiv, 2024 · paper
SciBERT: A Pretrained Language Model for Scientific Text — arXiv, 2019 · paper
SciMON: Scientific Inspiration Machines Optimized for Novelty — arXiv, 2023 · paper
SCITUNE: Aligning Large Language Models with Scientific Multimodal Instructions — arXiv, 2023 · paper
NatureLM: Deciphering the Language of Nature for Scientific Discovery — arXiv, 2025 · paper

Level 2: Agent As Partner

Level 2: Agent as Partner - multi-step collaboration and workflow orchestration

StarWhisper Telescope: Agent‑Based Observation Assistant System to Approach an AI Astrophysicist — arXiv, 2024 · paper
From Intention to Implementation: Automating Biomedical Research via LLMs — arXiv, 2024 · paper
CRISPR‑GPT: An LLM Agent for Automated Design of Gene‑Editing Experiments — arXiv, 2024 · paper
Towards an AI Co‑Scientist — arXiv, 2025 · paper
DrBioRight 2.0: An LLM‑Powered Bioinformatics Chatbot for Large‑Scale Cancer Functional Proteomics Analysis — Nature, 2025 · paper
ProtAgents: Protein Discovery via Large Language Model Multi‑Agent Collaboration — arXiv, 2024 · paper
MOOSE‑Chem: Large Language Models for Rediscovering Unseen Chemistry Scientific Hypotheses — arXiv, 2024 · paper
Autonomous Chemical Research with Large Language Models — Nature, 2023 · paper
ChemCrow: Augmenting Large‑Language Models with Chemistry Tools — arXiv, 2023 · paper
ORGANA: A Robotic Assistant for Automated Chemistry Experimentation and Characterization — arXiv, 2024 · paper
xChemAgents: Agentic AI for Explainable Quantum Chemistry — arXiv, 2025 · paper
ChemAgent: Self‑Updating Memories in Large Language Models Improves Chemical Reasoning — arXiv, 2025 · paper
Large Language Models to Accelerate Organic Chemistry Synthesis — arXiv, 2025 · paper
MyCrunchGPT: A ChatGPT‑Assisted Framework for Scientific Machine Learning — arXiv, 2023 · paper
MetaOpenFoam: An LLM‑Based Multi‑Agent Framework for CFD — arXiv, 2024 · paper
FoamAgent: Towards Automated Intelligent CFD Workflows — arXiv, 2025 · paper
AI‑Researcher: Autonomous Scientific Innovation — arXiv, 2025 · paper
Jupybara: Operationalizing a Design Space for Actionable Data Analysis and Storytelling with LLMs — arXiv, 2025 · paper
FlowAgent: Achieving Compliance and Flexibility for Workflow Agents — arXiv, 2025 · paper
The AI Scientist: Towards Fully Automated Open‑Ended Scientific Discovery — arXiv, 2024 · paper
GeoGPT: Understanding and Processing Geospatial Tasks through an Autonomous GPT — arXiv, 2023 · paper
PaperQA: Retrieval‑Augmented Generative Agent for Scientific Research — arXiv, 2023 · paper
ChatCite: LLM Agent with Human Workflow Guidance for Comparative Literature Summary — arXiv, 2024 · paper
PiFlow: Principle‑Aware Scientific Discovery with Multi‑Agent Collaboration — arXiv, 2025 · paper
Aviary: Training Language Agents on Challenging Scientific Tasks — arXiv, 2024 · paper
An Autonomous Laboratory for the Accelerated Synthesis of Novel Materials — Nature, 2023 · paper
Construction of a Knowledge Graph for Framework Material Enabled by Large Language Models and Its Application — npjCM, 2025 · paper
MultiCrossModal Automated Agent for Integrating Diverse Materials Science Data — arXiv, 2025 · paper
LLMatDesign: Autonomous Materials Discovery with Large Language Models — arXiv, 2024 · paper
Honeycomb: A Flexible LLM‑Based Agent System for Materials Science — arXiv, 2024 · paper
DrugAgent: Automating AI‑Aided Drug Discovery Programming through LLM Multi‑Agent Collaboration — arXiv, 2024 · paper
Toward a Team of AI‑Made Scientists for Scientific Discovery from Gene Expression Data — arXiv, 2024 · paper
MedAgents: Large Language Models as Collaborators for Zero‑Shot Medical Reasoning — arXiv, 2023 · paper
Automation of Systematic Reviews with Large Language Models — medRxiv, 2025 · paper
MRAgent: An LLM‑Based Automated Agent for Causal Knowledge Discovery in Disease via Mendelian Randomization — BriefBioinf, 2025 · website
GeneGPT: Augmenting Large Language Models with Domain Tools for Improved Access to Biomedical Information — arXiv, 2023 · paper
SR-Scientist: Scientific Equation Discovery With Agentic AI — ICLR, 2026 · paper · code
Fantastic Scientific Agents and How to Build Them: AgentBuild for Rietveld Refinement — arXiv, 2026 · paper
SciOrch: Learning to Orchestrate Expert LLMs for Solving Frontier Multimodal Scientific Reasoning Tasks — arXiv, 2026 · paper

Level 3: Agent As Avatar

Level 3: Agent as Avatar - autonomous research execution and long-horizon discovery

CycleResearcher: Improving Automated Research via Automated Review — arXiv, 2024 · paper
The AI Scientist‑v2: Workshop‑Level Automated Scientific Discovery via Agentic Tree Search — arXiv, 2025 · paper
AgentRxiv: Towards Collaborative Autonomous Research — arXiv, 2025 · paper
Agent Laboratory: Using LLM Agents as Research Assistants — arXiv, 2025 · paper
BiOMNI: A General‑Purpose Biomedical AI Agent — bioRxiv, 2025 · paper
OriGene: A Self‑Evolving Virtual Disease Biologist Automating Therapeutic Target Discovery — bioRxiv, 2025 · paper
CellVoyager: AI CompBio Agent Generates New Insights by Autonomously Analyzing Biological Data — bioRxiv, 2025 · paper
Sparks: Multi‑Agent Artificial Intelligence Model Discovers Protein Design Principles — arXiv, 2025 · paper
AlphaEvolve: A Coding Agent for Scientific and Algorithmic Discovery — arXiv, 2025 · paper
Robin: A Multi‑Agent System for Automating Scientific Discovery — arXiv, 2025 · paper
ScienceOne — Project, 2025 · website
EvoScientist: Towards Multi-Agent Evolving AI Scientists for End-to-End Scientific Discovery — arXiv, 2026 · paper · code
Self-Evolving Scientific Agent Discovers Generalizable Physically-Reasoned Fluid Control — arXiv, 2026 · paper

🧩 Method Guides

Guide	What it covers
Construction Methods	Knowledge organization, knowledge injection, tool integration, orchestration, and domain interfaces.
Enhancement Methods	Reasoning, memory, collaboration, workflow search, and self-review.
Evaluation Methods	Benchmark selection, executable evaluation, citation grounding, and long-horizon validation.
Auto-Research Systems	End-to-end research loops, coding/browser agents, training resources, and orchestration frameworks.

🧭 Broader Agentic & Auto-Research Repositories

Repository	Scope	Why follow it
karpathy/nanochat	Minimal end-to-end LLM training and chat stack	Tracks Karpathy's compact, hackable path from tokenizer and pretraining to finetuning, evaluation, inference, and chat UI.
karpathy/llm.c	LLM training in C/CUDA	Useful for understanding low-level training kernels, performance constraints, and reproducible GPT-style training.
SakanaAI/AI-Scientist	Automated idea-to-paper research loop	A reference point for autonomous ideation, coding, experimentation, paper writing, and automated review.
SakanaAI/AI-Scientist-v2	Agentic tree search for automated discovery	Follows the next iteration of AI Scientist with broader exploration and stronger end-to-end workflow design.
SamuelSchmidgall/AgentLaboratory	Human-guided autonomous research assistant	Shows how literature review, experimentation, and report writing can be composed into a full research workflow.
NoviScl/AI-Researcher	Research ideation and execution studies	Provides agent pipelines and human-study data for comparing LLM-generated ideas with expert research ideas.
OpenHands/OpenHands	AI software development agents	A generalist coding-agent platform for editing repositories, using terminals, browsing, and operating in sandboxed environments.
SWE-agent/SWE-agent	GitHub issue fixing and SWE-bench agents	A practical baseline for agentic software engineering, debugging, and repository-level task execution.
huggingface/smolagents	Lightweight code-agent framework	Good for studying minimal abstractions, code-as-action agents, sandboxed execution, and open-model agent workflows.
microsoft/autogen	Agentic AI programming framework	Useful for multi-agent conversations, orchestration patterns, and prototyping collaborative agent systems.
crewAIInc/crewAI	Multi-agent orchestration	Focuses on role-based agents, task delegation, crews, flows, and production-style automation workflows.
langchain-ai/langgraph	Graph-based agent workflows	Useful for durable, stateful, controllable agent graphs and long-running workflow orchestration.
FoundationAgents/MetaGPT	Multi-agent software company metaphor	A representative multi-agent framework for decomposing product/software work into role-specialized agents.
browser-use/browser-use	Browser automation for agents	Tracks web interaction patterns, browser control, and task automation over ordinary websites.
Significant-Gravitas/AutoGPT	Early autonomous agent platform	Still useful as historical context for goal-directed agents, autonomous task decomposition, and agent productization.
EvoScientist/EvoScientist	Self-evolving end-to-end AI scientist	Tracks persistent research memory, multi-agent experimentation, and human-on-the-loop research workflows.

✈️ Scientific Agents Construction

Agent construction methodology

Knowledge organization in scientific agents Orchestration and coordination in scientific-agent construction

Construction resources - prompts, context, tools, workflows, and domain interfaces

PromptAgent: Strategic Planning with Language Models Enables Expert-level Prompt Optimization — arXiv, 2023 · paper
Context Engineering: A Practical Handbook for Context Design, Orchestration, and Optimization — GitHub, 2025 · code
Verl: Volcano Engine Reinforcement Learning for Large Language Models — GitHub, 2024 · code
GraphRAG: A Modular Graph-Based Retrieval-Augmented Generation (RAG) System — arXiv, 2023 · paper
ChemCrow: Augmenting Large‑Language Models with Chemistry Tools — arXiv, 2023 · paper
Virtual Lab: AI Agents Design New SARS-CoV-2 Nanobodies with Experimental Validation — arXiv, 2024 · paper
OctoTools: An Agentic Framework with Extensible Tools for Complex Reasoning — arXiv, 2025 · paper
FoamAgent: Towards Automated Intelligent CFD Workflows — arXiv, 2025 · paper
MetaOpenFoam: An LLM-Based Multi-Agent Framework for CFD — arXiv, 2024 · paper
AutoPrompt: Eliciting Knowledge from Language Models with Automatically Generated Prompts — EMNLP, 2020 · paper
InstructBio: Instruction Tuning for Biomedical LLMs — arXiv, 2024 · paper
LangChain: Building Applications with LLMs through Composability — GitHub, 2022 · code
LightRAG: Simple and Fast Retrieval‑Augmented Generation — arXiv, 2024 · paper
SciTUNE: Aligning Large Language Models with Scientific Multimodal Instructions — arXiv, 2023 · paper
ClimateGPT: Towards AI Synthesizing Interdisciplinary Research on Climate Change — arXiv, 2024 · paper
SciMON: Scientific Inspiration Machines Optimized for Novelty — arXiv, 2023 · paper
TAIS: Gene Expression Agent with LLMs — arXiv, 2025 · paper
StarWhisper Telescope: Agent‑Based Observation Assistant System to Approach an AI Astrophysicist — arXiv, 2024 · paper
Fantastic Scientific Agents and How to Build Them: AgentBuild for Rietveld Refinement — arXiv, 2026 · paper

🚀 Scientific Agents Enhancement

Scientific-agent ability enhancement

Scientific agents' memory systems Illustration of scientific agent reasoning enhancement

Enhancement resources - reasoning, memory, workflow search, reflection, and collaboration

MemGPT: Toward LLMs as Operating Systems — arXiv, 2023 · paper
AFlow: Automating Agentic Workflow Generation — arXiv, 2023 · paper
ChemAgent: Self‑Updating Memories in Large Language Models Improves Chemical Reasoning — ICLR, 2025 · paper
DeepSeek‑Prover‑V2: Advancing Formal Mathematical Reasoning — arXiv, 2025 · paper
LLM‑SR: Scientific Equation Discovery via Programming with LLMs — ICLR, 2025 · paper
Sparks: Multi‑Agent Artificial Intelligence Model Discovers Protein Design Principles — arXiv, 2025 · paper
ether0: A Scientific Reasoning Model for Chemistry — arXiv, 2025 · paper
Robin: A Multi‑Agent System for Automating Scientific Discovery — arXiv, 2025 · paper
AgentRxiv: Towards Collaborative Autonomous Research — arXiv, 2025 · paper
ChatGPT Research Group for Optimizing the Crystallinity of MOFs and COFs — ACS, 2023 · paper
AI Co-Scientist: Towards an AI Co-Scientist — arXiv, 2025 · paper
ReAct: Synergizing Reasoning and Acting in Language Models — arXiv, 2022 · paper
Reflexion: Language Agents with Verbal Reinforcement Learning — NeurIPS, 2023 · paper
Self‑Refine: Iterative Self‑Improvement with Self‑Feedback — arXiv, 2023 · paper
Self‑Consistency: Reliable Decoding for Complex Reasoning — ICLR, 2023 · paper
AquilaChat: Agent with Long-Context Scratchpad Memory — arXiv, 2024 · paper
EvoScientist: Towards Multi-Agent Evolving AI Scientists for End-to-End Scientific Discovery — arXiv, 2026 · paper
SciOrch: Learning to Orchestrate Expert LLMs for Solving Frontier Multimodal Scientific Reasoning Tasks — arXiv, 2026 · paper
Self-Evolving Scientific Agent Discovers Generalizable Physically-Reasoned Fluid Control — arXiv, 2026 · paper

⚖️ Benchmark For Scientific Agents

Scientific-agent benchmark and evaluation landscape

Evaluation angle	Representative focus	Useful when you need to test...
Scientific knowledge and reasoning	BioMaze, SuperGPQA, Humanity's Last Exam, MR-Ben	Whether an agent can reason over expert-level scientific concepts.
Citation and literature grounding	CiteBench, ALCE, SurveyForge	Whether outputs are traceable, evidence-aware, and literature-faithful.
Code, data, and experiment execution	MLAgentBench, DSBench, SciCode, PaperBench	Whether an agent can implement, run, debug, and reproduce research workflows.
Domain and embodied environments	DiscoveryWorld, AgentClinic, GenoTEX, LLM-SRBench	Whether an agent performs in domain-specific or simulated scientific settings.

Benchmark resources — evaluation suites for scientific reasoning, data analysis, citation, coding, and agentic discovery

BioMaze: Benchmarking and Enhancing Large Language Models for Biological Pathway Reasoning — arXiv, 2025 · paper · dataset
BioKGBench: A Knowledge Graph Checking Benchmark of AI Agent for Biomedical Science — GitHub, 2025 · code
SurveyForge: On the Outline Heuristics, Memory-Driven Generation, and Multi-Dimensional Evaluation for Automated Survey Writing — arXiv, 2025 · paper · code
MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmark — NeurIPS, 2024 · paper · dataset
Humanity's Last Exam: A Hard Benchmark at the Frontier of Human Knowledge — arXiv, 2025 · paper · website
SuperGPQA: Scaling LLM Evaluation Across 285 Graduate Disciplines — arXiv, 2025 · paper · dataset
CiteBench: A Benchmark for Scientific Citation Text Generation — arXiv, 2022 · paper · code
ALCE: Enabling Large Language Models to Generate Text With Citations — arXiv, 2023 · paper · code
Tomato-Chem: Large Language Models for Rediscovering Unseen Chemistry Scientific Hypotheses — arXiv, 2024 · paper · code
Reviewer2: Optimizing Review Generation Through Prompt Generation — arXiv, 2024 · paper · code
Scientists' First Exam: Probing Cognitive Abilities of MLLM via Perception, Understanding, and Reasoning — arXiv, 2025 · dataset · paper
MR-Ben: A Comprehensive Meta-Reasoning Benchmark for Large Language Models — arXiv, 2024 · paper · website
FigureQA: An Annotated Figure Dataset for Visual Reasoning — arXiv, 2017 · paper · dataset
SciEval: A Multi-Level Large Language Model Evaluation Benchmark for Scientific Research — AAAI, 2024 · paper · dataset
MedXpertQA: Benchmarking Expert-Level Medical Reasoning and Understanding — arXiv, 2025 · paper · dataset
LLM-SRBench: A New Benchmark for Scientific Equation Discovery With Large Language Models — arXiv, 2025 · paper · dataset
GenoTEX: A Benchmark for Evaluating LLM-Based Exploration of Gene Expression Data in Alignment With Bioinformaticians — arXiv, 2024 · paper · dataset
MLAgentBench: Evaluating Language Agents on Machine Learning Experimentation — arXiv, 2023 · paper · code
PaperBench: Evaluating AI's Ability to Replicate AI Research — arXiv, 2025 · paper
DSCodeBench: A Realistic Benchmark for Data Science Code Generation — AAAI, 2026 · paper · code
DiscoveryWorld: A Virtual Environment for Developing and Evaluating Automated Scientific Discovery Agents — arXiv, 2024 · paper · code
SciCode: A Scientist-Curated Benchmark for Scientific Code Generation — arXiv, 2024 · paper · code
AgentClinic: A Multimodal Agent Benchmark to Evaluate AI in Simulated Clinical Environments — arXiv, 2024 · code · paper
SDRBench: Scientific Data Reduction Benchmark for Lossy Compressors — Website, 2021 · website · website
DSBench: How Far Are Data Science Agents from Becoming Data Science Experts? — GitHub, 2025 · code · paper
AISB: AI Scientist Benchmark — NLPCC, 2026 · evaluation kit
DSAgentBench: Can Agents Automate End-to-End Data-Science Workflows in Real Computer Environments? — arXiv, 2026 · paper · code

🤝 Contributing

🌞 Citation

@article{wang2025hitchhiker,
  title={The Hitchhiker's Guide to Autonomous Research: A Survey of Scientific Agents},
  author={Wang, Xinming and Xu, Jian and Feng, Aslan H and Chen, Yi and Guo, Haiyang and Zhu, Fei and Shao, Yuanqi and Ren, Minsi and Yi, Hongzhu and Lian, Sheng and others},
  year={2025}
}

Similar repos

OSU-NLP-Group/awesome-agents4science

A curated list of papers on LLMs and agents for scientific research and development

zoedsy/awesome-science-agents

AI agents for scientific discovery: papers, tools, and an interactive bilingual atlas with detailed…

du-nlp-lab/LLM4SR

LLM for Scientific Research Survey

133

ResearAI/Awesome-AI-Scientist

This is a survey of research on AI scientists, AI researchers, AI engineers, and a series of…

314

Omni-Scientist/Awesome-AI-Scientist

🧪 Awesome list of AI Scientist papers, systems, benchmarks, datasets and open-source platforms.

103

worldbench/awesome-ai-auto-research

🔥 A Survey on AI Auto-Research

533

ur-whitelab/LLMs-in-science

298

WuizaKaseiyo/awesome-deep-research-agent

638

Contributors

gudehhh666

27 commits

banjiuyufen

1 commits

Languages

Python

100.0%