Skip to content

How Agent D works

Agent D is the assistant built into Diagnify. It sits in the chat rail on the left of the console. You talk to it in plain words — or with short commands like /script, /labs, /refer, /bill, /plan and /recall — and it reads the chart, drafts work, and hands every act that matters to you on a Confirm card.

What it does

Agent D is one main agent with helper agents behind it:

  • The main agent is the one you talk to. It answers questions, drafts documents, and puts proposals on your Confirm card. It is the only part that can ask you to Confirm an act.
  • Ten helper agents cover the parts of the record: profile, diseases, medicines, investigations, preventive health, risk factors, vitals, care team, care gaps and billing. Every helper works from one shared patient summary built from the chart — the same facts for everyone. A field the record does not hold is named as missing, never guessed.

Helpers can read their own part of the record and draft suggestions. They cannot act. Only the main agent brings a proposal to you, and only you can press go.

When you ask for evidence, Agent D can run a deep search of public medical journals and show real paper citations — each one with a fixed ID such as [E12] that you can check. Licensed medicine-information sources are used behind the scenes but their names are never shown, on purpose.

With a patient pinned, a case card sits at the top of the chat: a short summary of who the patient is, their medicines, problems, vitals, last visit and open jobs.

When to use it

  • During a consult, to draft a script, a pathology or imaging request, a referral, a handout or the note — one at a time, or all at once with /plan.
  • Before the consult ends, to plan follow-ups and recalls (/recall), or to ask Prepare this visit and have the helpers check the whole chart for care gaps, preventive health, investigations and billing in one pass.
  • Any time, to ask a question about the record, a result, a guideline or the day's work — including from the Action list, so jobs you have already listed can be done the same way.

Step by step

Every request follows the same path:

  1. Ask. Type in the Agent D chat in plain words, use a slash command, or pick a job from the Action list.
  2. The helpers read the chart. They all work from the one shared patient summary, so they do not each invent their own version of the patient.
  3. One question form, if a fact is missing. If Agent D needs something the chart does not say, it shows one short question form — never two at once. Answer it, or skip it and type instead.
  4. One Confirm card. Everything the request produced is shown on one card — even a batch. A /plan puts the pathology, imaging, script, handout, referral and note on the same card. The card shows what will happen. Nothing has happened yet.
  5. You choose. Press go, or press No. No drops the whole proposal. Nothing is sent, signed or charged by Agent D itself.
  6. Done, with a note you co-sign. After your Confirm, Agent D writes an activity note under the consult note. Press Co-sign to finalise it, or Edit before co-sign to change it first — editing creates a new draft version.
flowchart TD
    A["You ask in chat, or pick from the Action list"] --> B["Helpers read the chart - one shared patient summary"]
    B --> C{"A fact is missing"}
    C -- "yes" --> D["One short question form"]
    C -- "no" --> E["One Confirm card"]
    D --> E
    E --> F{"You choose"}
    F -- "press go" --> G["The act is done - Agent D writes a note you co-sign"]
    F -- "press No" --> H["The proposal is dropped"]

What Agent D can do

  • Read the chart and the day's lists, and answer questions about them.
  • Draft scripts, requests, referrals, handouts, notes, recalls and invoices for you to check.
  • Run Prepare this visit and check the whole chart for gaps in one pass.
  • Watch for new results and patient messages, and put a draft plan in the Waiting on you tray.
  • Send a patient a secure request link by SMS so they can fill in a missing fact or a questionnaire — but only you press Send.

When Agent D opens another screen to show you something, your screen can follow it, or you can switch on Stay here to stay put.

Safety (what needs your Confirm)

Agent D never sends, signs, charges, claims or finalises on its own. Those are you-only acts — they always come to you on a Confirm card, and some (like sending a document or taking a payment) only happen inside the tool you open yourself. A script card that clashes with a recorded allergy keeps its go button off until you record an override and a reason.

Drafts stay drafts. Anything Agent D stages for a patient — a message, a recall booking, an order update — waits in Waiting on you and is not sent until you Confirm it.

Your practice also has a kill switch: when it is on, every write and send act is refused, whatever its card says. Reading and drafting still work. Ask whoever manages your clinic settings if you need it changed.

Every act — done or refused — is written to the audit log: who, what, when. See Safety and Confirm.

Common questions

Does Agent D read the whole chart? It works from one patient summary built from the record, and helpers read their own parts. It only sees the patient you have open or named.

Can Agent D message my patients by itself? No. It drafts messages and stages them in Waiting on you. You read them and press Send yourself. Appointment and recall reminder texts that your practice has already set up still go out on their hourly run.

What if I disagree with a draft? Press No on the Confirm card and the proposal is dropped. Tell Agent D what to change and it drafts again.

Are the citations real papers? Yes. Deep-evidence mode shows verified citations from public medical journals, each with a fixed ID like [E12] so you can find the paper. Licensed medicine-information sources stay unnamed.

Can it sit my CPD quiz? No. Type /cpd and it will tell you so.

Why did my screen just move? Agent D opened a screen to show you something. Your screen follows when it does. Switch on Stay here (shown next to the jump note) to stay where you are.

Last checked: 03/10/2026.