Awesome Healthcare Agents & Multi-Agents

A curated list of AI agents, multi-agent systems, benchmarks, datasets, RAG pipelines, safety tools, and research papers for healthcare and medicine.

Last updated: 2026-05-20

---

Overview

Healthcare agents combine large language models, retrieval, clinical tools, structured EHR access, medical imaging models, and human oversight to support tasks such as triage, clinical reasoning, documentation, drug safety, biomedical discovery, imaging analysis, and healthcare administration.

This repository focuses on resources that are useful for researchers and builders: open-source implementations, public benchmarks, reproducible datasets, and high-signal papers. It is not medical advice and should not be used to deploy autonomous clinical systems without clinical validation, governance, and regulatory review.

---

Table of Contents

• Agent Frameworks
• Healthcare Agent Systems
• Clinical Assistants & Medical LLMs
• Medical RAG & Knowledge Systems
• Imaging & Multimodal Agents
• Benchmarks & Evaluation Suites
• Open Datasets
• Safety, Privacy & Compliance
• Surveys & Position Papers
• Key Research Papers
• How to Contribute
• Citation
• License

---

Agent Frameworks

General-purpose frameworks that are useful for building healthcare-specific agents, tool-calling systems, and multi-agent workflows.

Project	Type	Why it matters for healthcare agents
AutoGen	Multi-agent framework	Conversational multi-agent orchestration, useful for specialist/critic/reviewer patterns.
CrewAI	Multi-agent framework	Role-based task delegation for operational and research workflows.
DSPy	Programming model	Optimizes LM pipelines and retrieval/programmatic modules.
Hugging Face smolagents	Lightweight agents	Minimal Python agent library with tool use and code agents.
LangChain	Agent/RAG framework	Common ecosystem for tool calling, retrieval, memory, and evaluation.
LangGraph	Stateful agent graphs	Good fit for auditable, branching clinical workflows with checkpoints.
LlamaIndex	Data/RAG framework	Connects documents, databases, vector stores, and agents.
Microsoft Semantic Kernel	Agent orchestration SDK	Enterprise-oriented skills/plugins, planners, memory, and workflow integration.
n8n	Workflow automation	Low-code automation for intake, notifications, scheduling, and back-office healthcare workflows.
OpenAI Agents SDK	Agent SDK	Lightweight Python SDK for tool calling, handoffs, tracing, and guardrails.
OpenAI AgentKit	Agent platform	Tooling for building, deploying, and optimizing production agents.

---

Healthcare Agent Systems

Research and open-source systems designed specifically for clinical, biomedical, or health-system agent workflows.

Project	Domain	Description
AgentMD	Clinical risk calculators	Language agent that curates and applies clinical calculators via the RiskCalcs toolset.
AI Hospital	Simulated clinical interaction	Multi-agent simulator with Doctor, Patient, and Examiner roles plus the MVME benchmark.
AgentClinic	Clinical simulation	Multimodal agent benchmark for sequential diagnosis, patient interaction, tool use, and data collection.
HealthAdminBench	Healthcare administration	Web-environment benchmark for evaluating agents on realistic administrative healthcare tasks.
MDAgents	Medical decision-making	Adaptive collaboration strategy that chooses solo, multidisciplinary, or integrated-care agent structures.
MEDDxAgent	Differential diagnosis	Modular agent framework with history taking, retrieval, and diagnostic strategy agents.
MedAgent-Pro	Multimodal diagnosis	Evidence-based agentic workflow for multimodal medical diagnosis with specialized tool agents.
MedAgentSim	Clinical simulation	Self-evolving doctor, patient, and measurement agents for dynamic diagnostic interactions.
MedAgents	Medical QA/reasoning	Multidisciplinary collaboration framework for zero-shot medical reasoning.
Meissa	Offline multimodal medical agents	Lightweight 4B medical agentic model trained on structured trajectories for radiology, pathology, and clinical reasoning.

---

Clinical Assistants & Medical LLMs

Medical LLMs and assistants that are frequently used as backbones or baselines for healthcare-agent research.

Project	Domain	Description
ChatDoctor	Medical consultation	LLM-based medical consultation model and dataset work.
Clinical Camel	Clinical reasoning	Open clinical LLM project trained for medical QA and reasoning.
ClinicalGPT	Clinical assistant	Fine-tuned clinical assistant built around EMR and note-style data.
HealthGPT	Personal health	Health-tracker assistant for longitudinal personal health data.
Med-Gemini	Medical foundation model	Google research model family for multimodal medical reasoning.
MedGemma	Open medical models	Google Health open models for medical text and multimodal use cases.
Med-PaLM 2	Medical reasoning	Google medical LLM evaluated on medical QA and clinical reasoning.
Meditron	Open medical LLM	Open Llama-derived medical LLM family for low-resource and clinical settings.
OpenBioLLM	Biomedical LLM	Open biomedical LLM family trained for biomedical and clinical QA.

---

Medical RAG & Knowledge Systems

RAG, graph-RAG, and biomedical knowledge systems for grounded medical answers.

Project	Type	Description
Baranzini Lab KG-RAG	Biomedical KG-RAG	Uses the SPOKE biomedical knowledge graph to provide prompt-aware context to LLMs.
MedGraphRAG / Medical-Graph-RAG	Graph RAG	Graph-based medical RAG for safer, evidence-linked medical responses.
MedRAG Toolkit	Medical RAG benchmark/toolkit	Toolkit for benchmarking retrieval-augmented generation in medicine.
MMed-RAG	Multimodal medical RAG	Multimodal RAG for medical vision-language models across radiology, ophthalmology, and pathology.
PubMed	Literature source	Core biomedical literature index for retrieval and evidence grounding.
UMLS	Terminology	Metathesaurus and controlled vocabularies for clinical concept normalization.

---

Imaging & Multimodal Agents

Medical imaging, report generation, segmentation, and multimodal tool-use projects.

Project	Domain	Description
BAAI Cardiac Agent	Cardiac imaging	Agentic system for cardiovascular magnetic resonance analysis.
BiomedCLIP	Vision-language model	Biomedical image-text foundation model often used as an imaging tool.
ChatCAD	Radiology assistant	LLM-based assistant for computer-aided diagnosis workflows.
MedRAX	Chest X-ray reasoning	Medical reasoning agent for chest X-ray interpretation and tool use.
MedSAM	Segmentation	Segment Anything adaptation for medical image segmentation.
MedSAM-Agent	Interactive segmentation	Agentic reinforcement-learning approach for multi-turn medical image segmentation.
MMedAgent	Multimodal tool agent	Medical agent that learns to select and use specialized multimodal tools.
MONAI	Imaging framework	PyTorch-based medical imaging framework for training and deployment.

---

Benchmarks & Evaluation Suites

Benchmarks for evaluating healthcare agents, clinical LLMs, RAG systems, and multimodal biomedical reasoning.

Benchmark	Domain	Description
AgentClinic	Interactive diagnosis	Multimodal simulated clinical environments spanning specialties and languages.
AgentRx	Multimodal clinical prediction	Benchmark study comparing single-agent and multi-agent systems on clinical prediction tasks.
AI Hospital / MVME	Clinical simulation	Multi-view evaluation of symptom collection, examination recommendations, and diagnosis.
BixBench	Computational biology agents	Real-world bioinformatics scenarios for long-horizon data analysis agents.
CalcQA / MeNTi	Medical calculator agents	Benchmark and nested-tool-calling architecture for medical calculator tasks.
ClinicalAgent Bench (CAB)	Clinical agents	18-task benchmark across realistic clinical dimensions from ReflecTool.
HealthAdminBench	Administration	Healthcare admin benchmark with live web portals and task-level scoring.
HealthBench	Health conversations	OpenAI benchmark for safety and usefulness in realistic health conversations.
MedAgentBench	Virtual EHR agents	Stanford virtual EHR environment for benchmarking medical LLM agents.
MedAgentGym	Biomedical coding agents	Training/evaluation environment with 72k+ code-based biomedical reasoning tasks.
MedAgentsBench