Making Vetrics usable through natural language

We have put OpenAI's agent harness inside Vetrics and connected it to the functions of the PMS.

That means a member of staff can open the assistant and ask Vetrics a question in ordinary language. The agent can use the patient record, client record, diary, rota, billing, stock, labs, imaging, inbox, insurance, CRM, documents, tasks, automations and the rest of the system to answer it.

We believe this is a first for veterinary software.

The interesting part is what happens once the agent can use the PMS rather than simply talk about it.

A practice manager can ask which clients owe more than £500, which patients are overdue vaccination, what revenue looked like by vet over the last six months, what is on the diary today, what is running low in stock or where the insurance claims pipeline is backing up.

Vetrics can answer with the same sort of data the user would normally find by moving around the PMS: records, lists, tables and charts.

The conversation can then continue. If you ask for revenue by vet and the first result is quarterly, you can tell Vetrics to make it monthly. Then exclude insurance pending. Then split surgery from consultations. Then save the final version and add it to the management dashboard.

You do not need to rebuild the query each time because Vetrics keeps the conversation and the objects created during it.

From a question to useful work

The aim is not simply to make data easier to find. The assistant can also use the functions behind Vetrics to prepare the next piece of work.

If a debtor list shows a group of clients worth chasing, you can ask Vetrics to create the tasks for reception or prepare the messages.

If you are looking at six images on a patient record, you can ask Vetrics to put them into a referral pack, a PDF or a presentation.

If you are on the rota and somebody is unavailable, you can ask who could cover the gap and have Vetrics prepare the change.

If you need to refer a case, Vetrics can gather the relevant history, results and images, prepare the referral document and draft the email to the referral contact.

The useful output stays inside Vetrics. Reports can be saved to Reporting or pinned to a dashboard. Documents can be attached to patients, clients or consultations. The same pack can be produced as a PDF, PowerPoint or spreadsheet without starting again.

Repeated processes can also be saved as practice skills. A clinic might save the way it prepares the Monday management pack, checks tomorrow's staffing, reviews overdue debt or prepares a referral. The next time, the same process is available to run again.

It knows what you are working on

The assistant is available inside the existing Vetrics interface, so it can use the context of the page you are already on.

On a patient record, "summarise the last six months" refers to that patient.

If six images are selected, "put these in a referral pack" refers to those six images.

On an invoice, "why is this so high?" can use the invoice and the records behind it.

On the rota, "who can cover this?" can use the shift, the staff record and the working pattern already held in Vetrics.

That matters because using AI inside a PMS should not mean repeatedly copying information into a separate box. The system already knows which patient, client, invoice, appointment or document you are working with.

Changes still belong to the user

We have drawn a hard line between looking at data and changing it.

Vetrics can search, calculate, analyse and visualise anything the signed-in user already has permission to see.

If a request would change the practice record, Vetrics prepares the action and shows it to the user before anything is written.

Ask it to move two appointments and it will prepare the moves.

Ask it to create a referral pack, attach it to the patient and draft the email, and it will show those actions together.

Ask it to refund a payment, change the rota, create a task or save a report, and the change waits for confirmation.

The confirmation is enforced by Vetrics itself. The agent cannot decide to skip it.

That lets us expose much more of the PMS through natural language without giving a model uncontrolled write access to practice data.

The same permissions still apply

The assistant works with the permissions of the person who is signed in.

A receptionist cannot ask the agent for information they cannot normally access. A manager who can work with reporting or rota can use those functions through the assistant. Practices can restrict access further by role or by individual user.

We have also designed the system so that AI providers receive less identifying information than the PMS itself holds.

Where the model only needs to know which patient or client is involved, Vetrics uses an internal ID rather than passing names, email addresses, telephone numbers or postal addresses.

Generated communications can use Vetrics merge fields. The model can write:

"Hi {{client.firstName}}, I'm writing about {{patient.name}} following today's consultation..."

Vetrics fills in the real names afterwards.

The same principle applies to files and images. If the job is to attach six images to an email or place them into a PowerPoint, the model does not need to inspect those images. Vetrics can work with their internal references.

If a clinician actually asks for visual analysis, that is handled as a separate, explicit operation.

The capability grows with Vetrics

There is a practical problem with most AI assistants built into software: they are given a small collection of hand-picked tools and quickly fall behind the product around them.

We are handling this differently.

Vetrics is building a catalogue of the actual capabilities in the PMS and connecting the assistant to those capabilities through the same services the normal interface uses.

When we add a new Vetrics feature, its natural-language capability is reviewed at the same time. Our development checks can flag parts of the product that have no assistant coverage.

That matters because Vetrics is already a large system. Patients, clients, consultations, appointments, rota, billing, reporting, stock, purchasing, labs, imaging, communications, insurance, CRM, documents, tasks, approvals, automations and other areas all contain useful functions.

The goal is that the assistant gradually becomes able to work across all of them rather than remaining a small AI corner of the product.

The PMS is still the PMS

A natural-language interface does not make screens redundant.

A week of appointments is still useful as a diary. A rota is still useful as a rota. An imaging study still belongs in an imaging viewer. Detailed billing work still belongs on the invoice.

The assistant gives staff a faster route into the work and a way to connect parts of the system that would otherwise require several screens and several manual steps.

You can still use Vetrics exactly as you do today.

You can also ask it for what you need.

For a product that contains thousands of records, dozens of workflows and a growing set of tools for running a practice, that changes how much of the system a member of staff can actually use.

If Vetrics knows how to do something, our aim is that you should be able to ask for it.