MediAgents Review
A LangGraph multi-agent Streamlit demo for symptom analysis, differential diagnosis, and treatment planning.
Each review covers deployability, value versus commercial alternatives, and privacy model. Tools that can run locally were started and exercised; mobile or backend-dependent tools were assessed from published builds, source code, and deploy guides. Ratings reflect what we were able to verify.
MediAgents — Open Source Multi-Agent Medical Diagnostics
A LangGraph multi-agent Streamlit demo for symptom analysis, differential diagnosis, and treatment planning. Last updated: 2026-06-21.
One-sentence verdict: A thoughtfully documented multi-agent diagnostics prototype with an evaluation suite; a research/educational project, not a validated clinical tool.
What the System Is
MediAgents (published as AI-Agents-for-Medical-Diagnostics) is a Python Streamlit app orchestrated by LangGraph 0.2.28. The deploy guide reports the service was started on port 5017. It routes a patient case through specialized agents and produces a structured report.
The documented pipeline:
- Patient Intake — demographics, symptoms, current medications, and history.
- Symptom Analyzer — web search-assisted symptom analysis with emergency detection.
- Diagnosis Agent — differential diagnosis reasoning.
- Treatment Agent — treatment suggestions with drug interaction checks via openFDA and a hardcoded critical table.
- Report Generator — structured clinical report.
The repository also includes an evaluation framework with ten synthetic test cases.
| Key data | |
|---|---|
| Category | Medical Diagnostics |
| Language | Python / Streamlit |
| License | MIT |
| Self-hosted | Yes |
| AI provider | Anthropic Claude (direct SDK) |
| Search / drug data | DuckDuckGo, openFDA |
| Memory | LangGraph MemorySaver + JSON patient histories |
How to Install and Deploy
cd /data2/docker/going-global/repos/AI-Agents-for-Medical-Diagnostics
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
export ANTHROPIC_API_KEY=sk-ant-...
streamlit run app.py --server.port 5017 --server.headless true
Then open http://localhost:5017.
The default login is admin / admin123. Actual diagnostic output requires a valid Anthropic API key.
There is no Docker Compose for this project.
How to Test
The documented test flow is:
- Open
http://localhost:5017and log in with the default credentials. - Enter a patient profile and symptom description in the intake form.
- Observe the live status updates as each agent runs.
- Verify that the final report includes sections for symptom analysis, diagnosis, and treatment.
- Try an emergency presentation (e.g., crushing chest pain) and confirm the pipeline routes to the emergency report.
- Run the evaluation suite with
python -m evaluation.eval_framework.
The README emphasizes that outputs are for research and education only.
Privacy & Compliance
MediAgents is not HIPAA compliant. Patient descriptions are sent to Anthropic's Claude API. Web search queries and drug lookups also leave the local machine. Session memory is in-process plus JSON files. Do not use it for real PHI or clinical decisions.
MediAgents vs Commercial Diagnostic Tools
| Dimension | MediAgents | Commercial Tools (e.g., Isabel Healthcare, Ada Health, Infermedica) |
|---|---|---|
| Cost | Free / self-hosted + API usage | Subscription / per-query |
| Data model | Cloud LLM + web search; local JSON history | Cloud-hosted, medically curated knowledge base |
| Clinical validation | None; synthetic test cases only | Medically reviewed, often regulatory-registered |
| Setup effort | Python venv + Anthropic key | SaaS signup or API contract |
| Explainability | Agent ledger in the UI | Varies by vendor |
| Customizability | Full source code | Limited |
Who Should Use It
- Researchers learning how to build multi-agent diagnostic pipelines.
- Teams evaluating LangGraph state machines for medical workflows.
- Developers who want a worked example of tool-using agents with openFDA integration.
Who Shouldn't Use It
- Clinicians making real diagnostic or treatment decisions.
- Organizations needing HIPAA-compliant or validated decision support.
- Users without an Anthropic API key who want a fully functional demo.
FAQ
Is MediAgents a medical device?
No. The README and UI explicitly state it is a research and educational project and must not be used for clinical decisions.
What LLM does it use?
It calls the Anthropic Claude API directly using the Anthropic Python SDK. You must provide an ANTHROPIC_API_KEY.
Does it check drug interactions?
The Treatment Agent queries the openFDA API and a hardcoded critical-interaction table. This is not a complete or validated drug-interaction checker.
Verdict
MediAgents is one of the more complete research prototypes in this batch. The multi-agent graph, live Streamlit status, openFDA integration, and evaluation framework show real engineering intent. It remains an educational artifact, not a clinical system.
Ratings: Deployability 4/5 · Value vs Commercial 2/5 · Privacy Compliance 2/5