We never wanted to build an EHR.

Patient data belongs to the practice and the patient. Not to the software vendor. Everything we build follows from that one sentence.

We did not want to build an EHR

PhiClaw started with a simple idea. Doctors wanted an AI employee that could run the front desk, answer the phones, follow up with patients, and keep the chart current, all under a signed Business Associate Agreement (BAA). We built that on top of OpenClaw, the open-source agent technology that can actually do work instead of just chatting. The plan was to plug it into whatever EHR the practice already used and get out of the way.

We had no interest in building an EHR. We did not want to compete with one. Practices have already paid for one, and switching records systems is one of the most dreaded projects in medicine. Our job was to add a worker to the practice, not to rip out its filing cabinet.

That plan did not survive contact with the big EHRs.

The vendors forced our hand

We asked major EHR vendors for an API so our agent could read and write the chart the practice already owned. The answer, over and over, was no. Some refused to give us an API at all. Some refused to allow any connection. Some simply stayed as closed as they could. The practice wanted it, the patient would have benefited, and the vendor still said no.

A practice pays a vendor to store its records. Then the practice asks to use its own records with a tool of its choosing, and the vendor decides the answer. That is not a security policy. That is control, and the doctor is the one paying for it.

So we built the open chart

If the chart would not open, we would build one that does. PhiClaw now ships with its own HIPAA EHR and CRM, built from the first line to be operated by an AI employee and by people side by side. Charting, records, pipelines, follow-ups, and e-prescribing (including controlled substances through our certified partner, with the licensed provider always reviewing and authorizing) live inside one compliant system. Every link in that chain, from the AWS and Amazon Bedrock infrastructure to the Convex database to the voice layer on calls, is covered by a BAA. You can read how that stack fits together on our about page.

We built it because somebody had to prove that a records system can be open without breaking HIPAA. Compliance is the foundation. Openness is the point.

The agent gets its own login

Most practices are not switching EHRs tomorrow, and we do not ask them to. So we taught the agent to work inside a closed EHR the same way a new human employee would. Where the vendor offers an API, FHIR, or HL7 connection, we use it, and today that covers more than 30 major EHRs directly or through our Keragon partnership. Where no API exists, the agent gets its own user account, its own computer, and its own screen. It logs in, finds the patient, reads the chart, and clicks the same buttons a front-desk coordinator clicks. Every action is logged and attributable, and the clinician stays the decision-maker on anything clinical.

A vendor can refuse an API. A vendor cannot refuse the practice the right to hire an employee and give that employee a login. That is why we can walk into almost any practice and start the same week. The details are in our guide to OpenClaw EHR integration and on the integrations page.

AI employees need an open chart

Here is the part the industry has not caught up to yet. A human employee can tolerate a bad system. They memorize the workarounds and keep a sticky note of which menu hides the referral form. An AI employee can do that too, and ours does, but it is a waste. The chart is the single most important input to every task in a practice, and when it is locked behind a front end that was designed for a mouse in 2009, every task gets slower, more expensive, and more likely to miss something.

So we are going one step further. We are shipping Marrow, a headless EHR: a records system with no front end at all. The chart is the API. There is no screen to click because there is nothing in the way. AI agents, custom apps, patient portals, and a practice's own tools all read and write the same structured record under the same BAA. If a practice wants a different interface, it can build one or buy one, and the data never moves.

Compliance is not an excuse for closure

The big EHRs often say no in the language of safety. We take HIPAA as seriously as anyone in this business. We sign a BAA with every practice before a single record is touched, we run on HIPAA-eligible infrastructure end to end, we minimize PHI before a model ever sees it, we encrypt in transit and at rest, and we log every action the agent takes. None of that required a closed chart. In fact, a chart you can audit through an API is easier to prove compliant than one you can only inspect by screenshot. Security is a reason to control access carefully, not a reason to deny it to the people who own the data. Our whitepaper walks through exactly how we draw that line.

What we owe the practices that trust us

We are a small team in Miami, started by a self-taught founder who shipped the first version in a couple of weeks because a doctor was waiting. We have no installed base to protect, and we intend to never become the thing we built against. The pricing is flat and month to month, migration in and out is free, and if you leave, your records go with you in a format another system can read. If a practice ever feels trapped with us, we have failed at the one thing we set out to do. If you want to help build this, we are hiring.

What we believe

1. Patient data belongs to the practice and the patient, never to the software vendor that stores it.
2. A chart you cannot get out of is a tax on care, paid every day by staff and doctors.
3. Refusing an API is a business decision dressed up as a security policy.
4. Compliance and openness are not in tension; a well-logged API is easier to prove compliant than a screen.
5. An AI employee should be able to do anything a human employee can do in a practice, including log in and click.
6. The best interface to a chart is no interface at all, just a clean, structured record any tool can use.
7. Migration in and migration out should both be free, because leaving should be as easy as joining.
8. The clinician makes the clinical decision; the software does the paperwork and never the judgment.
9. Any vendor, including us, that makes a practice feel trapped has stopped serving medicine.

We did not set out to change how medical records work. The EHR vendors left us no other way to do our job, so now we are. If this sounds like your practice, come meet the team.

John Gonzalez, founder