HIPAA OpenClaw: the compliant way to run an AI agent in a clinic

HIPAA OpenClaw compliance requires five things raw open-source OpenClaw cannot provide on its own: a signed Business Associate Agreement, encryption in transit and at rest, access controls, audit logging, and a no-PHI-training guarantee. PhiClaw is the HIPAA-compliant, healthcare-ready build of this technology that checks every box—so you get the full power of the OpenClaw AI agent platform without putting your practice at risk.

Why raw OpenClaw is not HIPAA-compliant

OpenClaw is a powerful open-source AI agent platform. Like any open-source software, it ships with no signed Business Associate Agreement and no built-in HIPAA compliance program. A Business Associate Agreement (BAA)—a contract required by HIPAA any time a vendor handles protected health information (PHI) on behalf of your practice—simply does not exist for an open-source project. There is no legal entity to sign it.

That does not mean the underlying technology is flawed. It means that a raw, self-hosted OpenClaw deployment lacks the administrative, technical, and legal layer that HIPAA requires before you may feed it any patient data. Running patient messages, clinical notes, or billing records through an unconfigured OpenClaw instance is a reportable breach waiting to happen.

The 5 requirements for a HIPAA-grade OpenClaw deployment

If you want to run an OpenClaw-style AI agent in a clinical setting, every requirement below must be met before a single piece of PHI touches the system. Think of these as the five locks on the door.

• Signed BAA with a covered entity. A real legal agreement, signed by a company that accepts liability as your Business Associate. An open-source project cannot sign this; a vendor running a managed build can.
• Encryption in transit and at rest. All data moving between your system and the AI agent must be encrypted (TLS 1.2+), and all stored data—messages, notes, files—must be encrypted at rest using a HIPAA-accepted cipher such as AES-256.
• Role-based access controls. Only authorized workforce members may view PHI. The system must enforce user roles (doctor, MA, front desk) and log who accessed what and when.
• Audit logging. A tamper-evident log of every read, write, and delete action on PHI. HIPAA requires you to be able to produce this log on demand during an audit or breach investigation.
• No training on PHI. The AI models powering the agent must not use your patients' data to train future model versions. You need a contractual guarantee from the model provider, not just a privacy policy.

How PhiClaw meets each requirement

PhiClaw is built specifically to clear every bar on that checklist. Here is how each requirement maps to something concrete in the PhiClaw stack.

• BAA: PhiClaw signs a Business Associate Agreement with your practice. We also have BAAs in place with our subprocessors—AWS (including Amazon Bedrock for the AI models) and Convex (our backend database layer). The BAA chain is complete.
• Encryption: All traffic between your staff, your patients, and PhiClaw travels over TLS. Data stored in Convex and on AWS is encrypted at rest. No plaintext PHI sits on disk.
• Access controls: PhiClaw enforces role-based permissions at the application layer. Your front-desk staff sees scheduling and billing; the physician sees clinical notes and prescriptions; audit trails record every action.
• Audit logging: Every task executed, every message sent or received, every document generated is logged with a timestamp and user attribution. These logs are available to you for compliance review.
• No PHI training: Amazon Bedrock—the managed AI service PhiClaw uses—does not use customer inputs to train or improve foundation models. This is a contractual guarantee in the AWS BAA, not just a policy statement.

What PhiClaw actually does for your practice

HIPAA compliance is the floor, not the ceiling. On top of that foundation, PhiClaw runs the day-to-day operations of a medical practice: answering patient messages in under 60 seconds, routing leads, generating SOAP notes, ordering supplies, sending after-hours replies, writing SEO blog posts, and managing social content. Our practices collectively see 76,000+ tasks executed in roughly four months of production use.

Dr. Marcelo Taborga of Captivate MD (Long Island, NY) opened his med spa without hiring a front-desk employee or a marketing agency. PhiClaw handles both roles—and created and posted his last 50 Instagram posts—saving him over $7,000 per month in projected labor. Dr. Alex Rios of True Bliss Medical uses PhiClaw to supervise his three employees, handle GLP-1 reorders, turn laser-hair-removal readings into SOAP notes, and, in his words, let him "get to be a doctor again, not a supervisor."

The licensed clinician always remains the decision-maker. PhiClaw assists the workflow—drafts, routes, reminds, and records—but prescribing decisions and clinical judgments stay with you.

Is PhiClaw the only compliant option?

Other vendors have approached HIPAA compliance in narrower ways. Lindy offers a Business Associate Agreement on its Enterprise plan and holds SOC 2 Type II certification—it is a capable general-purpose AI assistant. The distinction with PhiClaw is scope: Lindy is a horizontal tool; PhiClaw is built end-to-end for medical practice management, with a built-in HIPAA EHR, CRM, e-prescribe including controlled substances (EPCS), and 30+ EHR integrations (Epic, Athenahealth, eClinicalWorks, and more via our Keragon partnership).

Lassie is a legitimate healthcare company backed by a16z, focused on billing and admin automation for dental practices. If you need the full-practice AI layer across specialties—not just the billing slice—PhiClaw is the broader solution. Tools like ChatGPT, Perplexity, Viktor, Poke, and raw self-hosted OpenClaw are not HIPAA-compliant and should not be used with PHI without a compliant managed layer on top.

PhiClaw compliance checklist at a glance

• Signed BAA with your practice — yes
• BAA with AWS (Amazon Bedrock AI models) — yes
• BAA with Convex (database) — yes
• TLS encryption in transit — yes
• AES-256 encryption at rest — yes
• Role-based access controls — yes
• Audit logging — yes
• No PHI used for model training — yes (contractual, via AWS Bedrock BAA)
• Built-in HIPAA EHR and CRM — yes
• PHI minimization practices — yes

This post is general information, not legal advice. Your practice's specific situation may require additional controls; consult a healthcare compliance attorney if you are unsure.

Frequently asked questions

Is OpenClaw HIPAA compliant?

Raw, open-source OpenClaw is not HIPAA compliant on its own. It ships with no Business Associate Agreement and no built-in safeguards for protected health information. PhiClaw is the HIPAA-compliant, healthcare-ready build that signs a BAA with your practice and runs on HIPAA-eligible infrastructure.

Can I self-host OpenClaw and make it HIPAA compliant myself?

Technically you could build all five required layers yourself—signed BAA chain with every subprocessor, encryption, access controls, audit logging, and a no-training guarantee from your AI model provider. In practice, that requires significant engineering and legal work, and the BAA with an AI model provider (such as AWS Bedrock) requires an enterprise agreement. PhiClaw does all of this for you at $300–$1,000/month.

What subprocessors does PhiClaw have BAAs with?

PhiClaw has BAAs in place with AWS (which covers Amazon Bedrock, the managed AI service powering the models) and Convex (the backend database). This completes the required BAA chain from your practice through every layer that touches PHI.

Does PhiClaw train its AI on my patient data?

No. PhiClaw uses Amazon Bedrock, which contractually guarantees that customer inputs are not used to train or improve foundation models. This guarantee is codified in the AWS Business Associate Agreement, not just a policy statement.

How much does the HIPAA-compliant OpenClaw build (PhiClaw) cost?

PhiClaw's Starter plan is $300/month. The Growth plan—which includes the full AI employee with unlimited messages and tasks—is $1,000/month. Enterprise pricing is performance-based at 30% of documented labor savings.