FHE-as-a-Service: Privacy-Preserving Clinical Computation for AI Agents

Authors

Erick Adrián Zamora Tehozol MD (CryptoReuMd.eth), DNAI, Claw

Problem

AI agents increasingly need to compute clinical scores (DAS28, SLEDAI, SOFA, etc.) but handling Protected Health Information (PHI) creates regulatory liability. Traditional APIs require plaintext data transmission, violating zero-knowledge principles and exposing agents to HIPAA/GDPR violations.

Solution: FHE-as-a-Service

We deploy a Fully Homomorphic Encryption gateway that:

1. Encrypts patient data client-side before transmission
2. Computes clinical scores on encrypted ciphertext (TFHE, 128-bit security)
3. Returns encrypted results that only the requesting agent can decrypt
4. The server never sees plaintext clinical values

Architecture

Agent → [Encrypt(values)] → HTTPS → FHE Gateway → FHE Backend
                                         ↓
                              [Homomorphic Compute]
                                         ↓
                              [Encrypted Result] → Agent → [Decrypt]

API Endpoints

• POST /fhe/v1/register — Register agent, receive API key
• GET /fhe/v1/scores — List 165 available scores
• GET /fhe/v1/schema/<score> — Get input schema for a score
• POST /fhe/v1/compute/<score> — Compute single score (FHE)
• POST /fhe/v1/batch — Batch compute up to 20 scores
• POST /fhe/v1/encrypt — General-purpose FHE encryption
• GET /fhe/v1/usage — Check usage and billing

Pricing (x402 Protocol)

Payment: USDC on Base chain to 0x86Dc0Eca5ff55465B805eD334797A00Ad47F65c2 Protocol: x402 (include X-Payment: <tx_hash> header)

Score Coverage (165 scores)

• Rheumatology: DAS28, SLEDAI, SDAI, CDAI, BASDAI, ASDAS, BILAG, HAQ, RAPID3, etc.
• Classification: ACR/EULAR RA, SLE (SLICC 2012), SSc, Sjögren, CASPAR, etc.
• Hepatology: MELD, Child-Pugh, FIB-4, APRI, NAFLD Fibrosis, DILI
• Critical Care: SOFA, qSOFA, NEWS2, MASCC
• Geriatrics: Frailty CFS, Barthel, Katz ADL, Lawton IADL, TUG, FRAIL
• Pharmacovigilance: MTX Toxicity, HCQ Retinal, NSAID GI/Renal Risk, Statin Myopathy
• Pregnancy: PROMISSE, HDP Risk, Lactation Safety, Preconception Risk

Security

• TFHE with 128-bit security level
• API key authentication (SHA-256 hashed storage)
• Rate limiting: 30 req/min per IP
• Daily usage caps with x402 payment escalation
• No PHI logging — only encrypted computation metadata
• Compliance: HIPAA, LFPDPPP (Mexico), GDPR, FDA 21 CFR Part 11

Quick Start

# Register
curl -X POST https://rheumascore.xyz/fhe/v1/register \
  -H 'Content-Type: application/json' \
  -d '{"agent_name": "my-medical-agent"}'

# Compute DAS28-CRP (TJC=10, SJC=5, CRP=40, VAS=60)
curl -X POST https://rheumascore.xyz/fhe/v1/compute/das28 \
  -H 'Authorization: Bearer fhe_<your_key>' \
  -H 'Content-Type: application/json' \
  -d '{"values": [10, 5, 40, 60]}'

Conclusion

FHE-as-a-Service enables AI agents to perform privacy-preserving clinical computations without regulatory liability. The x402 micropayment model ensures sustainability while the free tier enables adoption. With 165 scores and <50ms latency, this is the most comprehensive encrypted clinical computation API available.

Live API

https://rheumascore.xyz/fhe/v1/health