The AI layer for the 49% of physician time that doesn't involve patients.
Ambient AI scribes like Abridge ($1B raised) and Nabla (100,000+ clinicians) solve one problem — the 27% of physician time that involves patients.
Nobody built AI for the other 49%.
"Physicians spend 49% of their time on administrative tasks. Only 27% with patients."
— NEJM Catalyst
"Prior authorization denials cost US hospitals $19.7 billion annually."
— AMA 2023
"The average discharge summary takes 45 minutes to write. 60% are never read."
— JAMA
The prior authorizations. The discharge summaries. The insurance appeals. The referral letters. The medication reconciliations.
This is where Nota Health lives.
|
Chat with any clinical document in plain language. Upload a discharge summary, prior auth, lab report, or referral letter and ask questions about it. AI responses stream in real time using your own API key. |
Pull structured data from clinical documents automatically. Medications with dosages, diagnoses with ICD-10 codes, lab values with reference ranges, follow-up actions — all extracted into clean tables. |
|
Pre-built AI workflows for the most common clinical documentation tasks: Prior authorization review, discharge summary analysis, referral letter drafting, insurance appeal generation, SOAP note formatting. |
Four standalone tools powered by free public APIs — no AI key required: Drug database and interaction checker, PubMed literature search, US provider verification (NPI Registry), ICD-10 code lookup. |
Nota Health connects to six free public clinical APIs out of the box. No commercial dependencies. No API keys required for clinical tools.
| Integration | What It Provides | Auth Required |
|---|---|---|
| OpenFDA | Drug labels, warnings, adverse events, recalls | No |
| RxNav (NLM) | Drug interactions, RxNorm normalization | No |
| NPI Registry (CMS) | US provider verification and lookup | No |
| PubMed (NCBI) | Clinical literature search and citation verification | Optional |
| ICD-10 (CMS) | Diagnosis code lookup for documentation and billing | No |
| ClinicalTrials.gov | Trial protocols and eligibility criteria | No |
| Metric | Source |
|---|---|
| 49% of physician time goes to admin, only 27% with patients | NEJM Catalyst |
| 15.6 hours/week spent on documentation by clinicians | Annals of Internal Medicine |
| $19.7B/year cost of prior auth denials to US hospitals | AMA, 2023 |
| 1 in 5 prior auths initially denied, 75% approved on appeal | KFF |
| Nota Health | Typical SaaS vendor | |
|---|---|---|
| Where your data lives | Your own Supabase project, your own infrastructure | Vendor's servers |
| Source code | Fully open, AGPL — read every line | Closed |
| Pricing | Free to self-host, forever | Per-seat subscription, usually annual |
| AI provider | Bring your own key — Claude, GPT, or Gemini | Locked to whatever the vendor chose |
| Customization | Fork it, extend it, change anything | Feature requests go into a backlog you don't control |
| Exit | Your data, your database — leave anytime | Contract renewal, data export process, migration risk |
Nota Health doesn't compete on having a bigger team or a sales process. It competes on not needing either.
Good fit if you:
- Run a clinic or hospital and want AI on your documents without a permanent vendor contract
- Have a Supabase project and an Anthropic/OpenAI/Gemini API key
- Want to self-host and control exactly where patient data lives
Not a fit if you:
- Need a plug-and-play SaaS with a support team on call — this is open source, you run it
- Need in-visit ambient transcription — Nota Health starts after the encounter, not during it
- Are not comfortable deploying and maintaining a self-hosted app
 |
 |
| Clinical Assistant — chat with any medical document | Clinical Extract — structured data from clinical documents |
 |
 |
| Drug Database — real FDA data, interaction checking | Provider Verification — live NPI Registry lookup |
Go to https://nota-health-hub-ai.lovable.app/
Sign up, add your own AI API key in Settings, and start uploading documents. No installation required. Your documents never leave your Supabase instance.
For clinical environments where patient data must stay on your own infrastructure.
Prerequisites:
- A Supabase project (free at supabase.com)
- At least one AI API key: Anthropic, OpenAI, or Google Gemini
Steps:
# 1. Fork this repository on GitHub
# 2. Create a Supabase project at supabase.com
# 3. Deploy to any platform that supports Vite/React apps
# (Vercel, Netlify, Railway, Lovable, Cloudflare Pages, or your own server)
# - Import the forked repo
# - Connect your Supabase project
# - Set the environment variables below
# Set these environment variables:
# VITE_SUPABASE_URL=your-supabase-url
# VITE_SUPABASE_ANON_KEY=your-supabase-anon-keyEnvironment Variables:
VITE_SUPABASE_URL=https://your-project.supabase.co
VITE_SUPABASE_ANON_KEY=your-anon-keyAI API keys are stored per-user in Supabase — not in environment variables. Each user adds their own key in Settings.
┌─────────────────────────────────────────────┐
│ Frontend (React + Vite) │
│ │
│ Clinical Assistant │ Cases │ Extract │
│ Protocols │ Drug DB │ PubMed │ Provider │
└──────────────────────┬──────────────────────┘
│
┌──────────────────────▼──────────────────────┐
│ Supabase │
│ │
│ Auth │ Postgres │ Storage │ RLS │
│ │
│ Tables: cases, documents, conversations, │
│ messages, extractions, audit_logs │
└──────────────────────┬──────────────────────┘
│
┌──────────────────────▼──────────────────────┐
│ Clinical APIs (free, public) │
│ │
│ OpenFDA · RxNav · NPI Registry │
│ PubMed · ICD-10 · ClinicalTrials.gov │
└─────────────────────────────────────────────┘
│
┌──────────────────────▼──────────────────────┐
│ AI Providers (user's own key) │
│ │
│ Anthropic Claude · OpenAI · Google Gemini │
└─────────────────────────────────────────────┘
Key architecture decisions:
- Bring your own AI key — Nota Health never holds your API keys. Each user adds their own key, stored encrypted in Supabase.
- Row-level security — Every Supabase table enforces RLS. No user can access another user's data, even if the application layer is compromised.
- Self-hostable storage — Documents are stored in Supabase Storage. Self-hosted deployments keep all patient data on your own infrastructure.
- No vendor lock-in — Works with Anthropic, OpenAI, or Gemini. Switch providers in Settings.
| Control | Status |
|---|---|
| End-to-end encrypted document storage | ✅ |
| Row-level database security (no cross-user access) | ✅ |
| Encrypted API key storage | ✅ |
| Complete audit log (action + timestamp, no PHI) | ✅ |
| Multi-factor authentication | ✅ |
| Automatic session timeout | ✅ |
| Full data export on request | ✅ |
| Permanent data deletion | ✅ |
| Self-hostable on any infrastructure | ✅ |
| Open source — full code auditability | ✅ |
| Business Associate Agreement (BAA) | ○ Self-hosted enterprise |
| SOC 2 Type II | ○ Roadmap |
On HIPAA: Nota Health is not a certified HIPAA-compliant service. It is architected with HIPAA principles in mind. For regulated clinical environments, deploy self-hosted and execute a BAA with your cloud provider.
Nota Health ships with 8 pre-built AI protocols for common clinical documentation workflows:
Assistant Protocols (conversational):
- Prior Authorization Review — identifies denial risks and missing documentation
- Discharge Summary Analysis — extracts key clinical information
- Referral Letter Draft — structures specialist referrals
- Insurance Appeal Letter — generates appeals with clinical justification
- SOAP Note Formatter — structures clinical notes
Extraction Protocols (structured table output):
- Medication List — drug name, dose, frequency, route, prescriber
- Diagnosis Summary — diagnoses with ICD-10 codes
- Lab Results — values with reference ranges and abnormal flags
v1.0 — Current
- ✅ Clinical document chat (multi-model)
- ✅ Structured clinical extraction (6 protocols)
- ✅ Drug database and interaction checker
- ✅ PubMed literature search and citation verification
- ✅ US provider verification (NPI Registry)
- ✅ ICD-10 code lookup
- ✅ Prior authorization review
- ✅ Insurance appeal generation
- ✅ Audit logging
- ✅ MFA support
v1.1 — Next
- Team workspaces with role-based access
- FHIR R4 document export
- Custom extraction schemas
- Bulk document processing
- MCP connector support — provider-agnostic tool use for the Clinical Assistant
v2.0 — Future
- EHR webhook integration
- SOC 2 Type II certification
- On-premise Docker deployment
Nota Health is open source under AGPL-3.0 and welcomes contributions.
Ways to contribute:
- Report bugs via GitHub Issues
- Suggest clinical workflows that should be built-in protocols
- Add extraction templates for your clinical specialty
- Improve documentation
- Star the repo — it helps more clinicians find Nota Health
To contribute code:
- Fork the repository
- Create a feature branch
- Make your changes
- Open a pull request with a clear description
| Layer | Technology |
|---|---|
| Frontend | React, TypeScript, Vite |
| UI | Tailwind CSS |
| Database | Supabase (Postgres) |
| Auth | Supabase Auth |
| Storage | Supabase Storage |
| Hosting | Lovable / any static host |
| AI | Anthropic Claude, OpenAI, Google Gemini |
| Clinical APIs | OpenFDA, RxNav, NPI Registry, PubMed, ICD-10 |
Nota Health is licensed under AGPL-3.0.
This means:
- ✅ Free to use for any purpose
- ✅ Free to self-host for your team or organisation
- ✅ Free to modify and extend
- ✅ Commercial use permitted
⚠️ If you distribute a modified version, you must open-source it under AGPL-3.0
Pavan Galiveeti — Lead PM building AI products for enterprise teams. Built with Claude Code as AI pair programmer. Building in public. All feedback welcome.
- LinkedIn: linkedin.com/in/pavangaliveeti
- GitHub: github.com/gpavan1992
Nota Health is not a medical device.
All AI outputs require review by a qualified healthcare professional.