I've been thinking about what personal software could mean for healthcare.

Imagine being able to describe a small tool you need and have an AI help you build it. Something that arranges your medical history in a way you understand. A dashboard that follows your lab results over time. An application that helps you prepare questions before an appointment.

You might use it for years. You might replace it next month.

But where does it get your medical history?

That question keeps bringing me back to what we are building with Merdra.

You can build a beautiful interface and still spend hours searching for old results, uploading documents, and explaining the same history again. If your information is scattered across hospitals, laboratories, paper files, and messages, creating another application does not automatically bring it together.

The software might be personal. The data it needs is still somewhere else.

A future of personal software would make this problem more visible. If more people can create the tools they want, more of those tools will need a reliable way to work with information that already exists.

Consider a simple example.

You want an application that displays your lab results over time. You give it permission to retrieve the relevant records, and it builds a timeline. Each result includes when the sample was collected, the unit, the reference range supplied by the lab, and where the result came from.

A few months later, you decide you want a different interface. Perhaps you prefer a local application on your laptop. Perhaps you want an agent to help organise your history before an appointment.

You should be able to make that change without reconstructing your medical history from scratch.

But why would anyone want to do this?

Honestly, building and maintaining your own health software sounds like a lot of work. Most people probably want something that already works. They do not want another project to manage.

Still, I can imagine cases where a person's needs are specific enough to make it worthwhile. Someone might want a simple timeline of results from different laboratories. Another might want an accessible interface in their preferred language. Someone managing a long-term condition might want to organise information around the questions they regularly discuss with their doctor.

They would not necessarily build these tools themselves.

This is the direction I want us to prepare Merdra for: patient-data infrastructure that different applications can work with, under appropriate permissions.

We cannot predict every useful piece of software someone will want. I don't think we need to. We need to think carefully about what those tools should be able to rely on.

One of those things is the source of a record.

A result issued by a laboratory, a symptom someone enters themselves, and an explanation generated by an AI are different kinds of information. A useful system should preserve those differences as information moves between applications.

If an agent produces a summary, someone should be able to trace it back to the records it used. If a laboratory corrects a result, applications should have a way to discover that correction. An old copy should not quietly appear to be the latest version.

Making information easier to move also means being more precise about permission.

Someone might want to let an application read their lab results without giving it access to every other part of their medical history. They might allow access for a week, or allow it to receive new results until they turn that permission off.

An application acting on a patient's behalf should only receive information that patient is permitted to access and authorised to share.

An agent acting for a patient should have a defined scope of access. It should be possible to see what it accessed and stop future access. That still leaves a difficult question about copies: revoking permission cannot magically erase information an external application has already downloaded.

These details become part of the product. They shape whether people can trust the software they connect.

Then there is security.

Every additional application that receives medical information creates another place where that information could be exposed. Patient permission alone would not make the connection safe.

We would need to consider how applications identify themselves, how little information they actually need, how access is recorded, and how it can be stopped. Patients should be able to understand what they are granting without needing to understand the infrastructure underneath it.

There is another boundary worth keeping clear: access to trustworthy records does not make an application clinically trustworthy.

An application can retrieve an authentic result and interpret it badly. An agent can cite the right document and still reach the wrong conclusion. Merdra's infrastructure would need to preserve evidence and accountability, while the software built on top would still need evaluation appropriate to what it does.

That matters to how I think about our own role.

We can build patient interfaces and explore medical agents. But I don't want the usefulness of the underlying infrastructure to depend on everyone choosing our interface.

Someone else may build a better experience for a particular person, language, or need. A patient may prefer a simpler tool. A developer may create something we would never have considered.

I want those possibilities to have somewhere reliable to start.

A useful early test would be whether an independently built application can connect with permission, retrieve results, preserve their context, and respond when something changes. It should also stop retrieving data when its access is revoked.

That is a smaller problem than supporting every imaginable medical agent, but it would tell us something real about the foundation we are building.

How much of this becomes part of everyday healthcare remains to be seen. Some people will keep choosing established products. Others may want tools built around their own routines.

For now, I'm just imagining what that future might look like.