How Eugene Vestel Built a Personal Health Agent with OpenClaw
Built by Anonymous · · Source: the web
Result: Vestel says the agent helped them understand a colonoscopy bill that triggered a $1,000 deductible, draft a dispute and decide whether to escalate. The source does not say how the dispute ended.
What they built
Eugene Vestel built a personal health agent on OpenClaw, paired with Claude, that works from their own medical and insurance records. According to Vestel, it prepares for appointments, drafts follow-up messages, fills out forms, writes insurance appeals, tracks long-term trends and flags cost issues. They say the goal is not to replace doctors.
The source does not mention WhatsApp or replacing a set of apps, which the earlier card on this site claimed. The write-up is about healthcare only.
How it works
Vestel describes two main pieces:
- A guardrail proxy. Vestel built this first and says they generated it with Claude Code. It sits between the agent and any FHIR server, and every request passes through it. It masks, truncates and strips identifying health information, keeps immutable audit logs, requires step-up authorization for writes, asks a human to confirm clinical actions, keeps tenants isolated and adds medical disclaimers when data is read. The code is linked on GitHub as the MCP Guardrails Project.
- A personal FHIR server. A private HAPI FHIR instance stores and normalizes the data. Vestel says it holds Epic patient-access records from two health systems, clinical records not available through TEFCA, pharmacy data, insurance coverage and claims pulled through Flexpa, and family data that is scoped and kept separate.
Vestel also lists Josh Mandel's Health Skillz repository as a way to connect to and download Epic data. The article does not give step-by-step setup instructions.
Vestel calls OpenClaw unsafe by default and not ready for consumers, and names prompt injection, skill misuse and over-permissioning as risks. The rules they say they follow:
- Never install random skills.
- Never give the agent direct access to raw health data.
- Run agents on isolated machines. Vestel's Mac mini has its own account.
- Reduce sensitive data at the data layer, not the model layer.
- Require human confirmation for clinical or financial actions.
The result
Vestel gives one concrete example. A colonoscopy was billed as diagnostic instead of routine, which created a $1,000 deductible. They say the agent helped them understand the billing difference, draft a dispute, find points of pressure such as credit reporting thresholds, and decide whether escalating was worth it. The article does not say whether the dispute was resolved or any money was recovered.
How to build something similar
These steps are our general advice, not Vestel's.
- Start by collecting your own records in one place. Many health systems and insurers offer patient-access exports, and a private FHIR server is one way to store them.
- Put a filtering layer between the agent and the data before you connect anything. Decide what gets masked, what gets logged and which actions need your approval.
- Run the agent on a separate machine or user account with only the access it needs, and only add skills you have read and trust.
- Begin with read-only tasks such as appointment prep or explaining a bill, and review everything it drafts before it goes to a provider or insurer.
The original post
Want to build something like this?
Join the Shipping Skool community free. Paid members get live bootcamp calls every week.
Join Shipping Skool — FreeMore use cases
- How Vinay Moved an AI Second Brain from Claude Code to Grok BotPersonal Productivity
- How Evan Kirstel Uses Claude Cowork as a Help Desk for Mac and iPhonePersonal Productivity
- How Khe Hy Built an Ambition-Upgrade Skill for Prompts and TasksPersonal Productivity
- How Sherwood Uses Hermes as a Personal Agent for Inbox TriagePersonal Productivity