AGx builds clinical agents that monitor the information around a patient, draft the next action, and bring it to the clinician. The clinician keeps every decision. The agent does the legwork.
An agent isn't a chat window — it's a presence that keeps working while you're with the patient. AGx is the face of that idea: always on, doing its rounds, ready to hand you what it found.
The steps a clinician needs aren't the hard part — good prompting can reason through any of them. What a chatbot can't do is start on its own, touch the systems around the patient, and keep going until the outcome is reached. That residue is the only thing that's actually agentic.
You ask, it answers, the job ends. Every step needs you to prompt the next one. It reasons well — but it never acts, and it forgets the moment the reply is sent.
It watches for a trigger, decides its own path through the work, drafts the action, and hands you the decision — then goes back to watching. The clinician didn't open a chat. The agent noticed.
A new lab lands, a prescription changes — the agent begins the work on its own. You can't prompt your way to initiative.
It reads records, checks databases, drafts notes, routes alerts. The output isn't a message — it's a changed state in a real system.
Its job ends when the outcome is reached, not when it replies. It can return tomorrow, see the recheck didn't happen, and raise it again.
An AGx agent isn't a fresh model guessing at clinical questions. It sits on top of the tools already built and validated in the ecosystem — the genomics engines and the CDx decision support — and adds one new layer: the ability to run a loop, decide a path, and act. The reasoning underneath is work that already exists.
Orchestration & action — the new layer
Clinical decision support — the diagnostic engine
The validated base — five genomics engines
The console brings clinical context, agent routing, live chain status, handoff summaries, stop control, copy/print tools, and clinician gates into one command surface across the full patient journey.
The clinician types, dictates, attaches readable files, chooses a trigger, or runs a single agent manually. This is the entry point for the workflow, not a final instruction to act.
AGx extracts the useful clinical facts, detects missing information, selects the relevant agents, and explains why each agent was queued. If the worker is unavailable, local fallback routing still prepares a safe draft route.
After context analysis, the cockpit displays detected facts, missing prompts, the agent routing board, a plain-language handoff summary, and a mission log. Internal payloads stay hidden from the user.
Each queued agent calls the tools relevant to its task: preparation checks, medication safety, CV risk, CDx reasoning, lab follow-up, referral workup, monitoring, recall, reconciliation, or patient communication.
AGx converts tool results into a structured draft. Copy-plan buttons, the chain summary, patient journey card, and print view help the clinician review what was produced without exposing technical internals.
Every draft stops for review. The clinician approves, edits, or rejects. AGx can organize the information work, but it never signs, sends, prescribes, orders, or acts without clinician sign-off.
Automation can continue to the next queued agent, the Stop button halts the chain after the active step, and the console refresh clears the cockpit for a new clinical context.
No route detected yet.
Enter a patient's demographics, problem list, and history. AGx-01 checks guideline-based screenings, vaccinations, and care gaps — then drafts the brief you read before walking in.
Enter demographics and history. Nothing is stored.
Guideline lookups, gap identification, and the drafted brief.
This is a demonstration of guideline-based pre-visit preparation. All recommendations must be verified by a licensed clinician against the actual patient record.
Enter a medication list and minimal context. AGx-02 checks interactions, PGx risk, renal dosing, and appropriateness — then drafts a medication-safety plan. You see every step in the trace, then you decide.
Nothing is stored. The agent runs server-side; tools produce the facts.
Every reasoning step and tool call, in order — so you verify the agent, not trust it.
Nothing here reaches a patient or a record until you act. The agent drafts and proposes; the clinical decision is yours. This is a demonstration tool, not a medical device, and its output must be verified by a licensed clinician.
Enter lipids, blood pressure, and risk factors. AGx-03 runs the Pooled Cohort Equation, checks statin eligibility against ACC/AHA guidelines, assesses your BP target, and drafts the risk conversation.
Enter demographics, lipids, and risk factors.
ASCVD calculation, statin eligibility, and BP assessment.
This is a demonstration tool. The Pooled Cohort Equation and statin guidelines are implemented for educational purposes. All risk assessments must be verified by a licensed clinician.
Enter a clinical presentation. AGx-04 rebuilds CDx as an agent with tools: validated scoring systems, diagnostic criteria, and risk stratification in one traceable loop.
Describe the case and select scoring systems.
Scoring calculations, criteria checks, and risk stratification.
Clinical scoring is an adjunct — scores do not replace clinical judgment. All assessments must be verified by a licensed clinician.
Enter a flagged lab result and patient context. AGx-05 interprets the value against reference ranges, identifies probable causes from the patient's conditions and medications, and drafts the follow-up with a plain-language patient message.
Enter the abnormal result and patient context.
Interpretation, differentials, and the drafted follow-up plan.
This is a demonstration tool. All lab interpretations and recommendations must be verified by a licensed clinician against the actual patient record and clinical context.
Enter the target specialty and clinical context. AGx-06 checks the expected workup, assesses urgency against red flags, and drafts the specialist letter.
Specify the specialty and reason.
Workup checklist, urgency assessment, and drafted referral.
Referral requirements vary by specialist and institution. Verify against local protocols.
Enter chronic conditions and current values. AGx-07 tracks disease targets, flags anything off-goal, and suggests escalation for clinician review.
Enter conditions and latest values.
Target checks, adequacy assessments, and escalation pathways.
Targets are guideline-based and must be individualised by the clinician for each patient.
Enter a patient's conditions and last-known dates. AGx-08 finds overdue rechecks, prioritises them, and drafts reminder messages.
Enter conditions and history.
Overdue items, priority ranking, and drafted recall message.
Recall intervals are guideline-based and should be verified against the patient's actual clinical history.
Paste the admission and discharge medication lists. AGx-09 compares them side by side, flags every change, checks new interactions, and drafts patient instructions.
Enter both medication lists.
Medication comparison, flagged changes, and interaction check.
Medication reconciliation must be verified by the clinician against the actual discharge summary and patient record.
Paste a clinical plan or diagnosis. AGx-10 turns it into plain-language explanations, supports translation-ready patient communication, and verifies readability with Flesch-Kincaid scoring.
Paste the clinical plan for the patient.
Term simplification, plan translation, and readability check.
Patient communications must be reviewed by the clinician before delivery.
Each agent owns one clinical problem end-to-end. Before, during, after, and between visits — the complete loop.
Reviews history, surfaces overdue screenings and care gaps, drafts a 30-second brief.
Live demoChecks interactions, PGx risk, renal dosing, and appropriateness.
Live demoCalculates 10-year ASCVD risk, checks statin eligibility, assesses BP targets.
Live demoCDx rebuilt as an agent with tools for scoring, criteria checks, and risk stratification.
Live demoInterprets flagged results, proposes actions, drafts patient communication.
Live demoChecks expected workup, assesses urgency, drafts the specialist letter.
Live demoTracks targets, flags when off-goal, suggests escalation.
Live demoFinds overdue rechecks, prioritises recalls, drafts reminders.
Live demoCompares admission vs discharge meds, flags changes, checks new interactions.
Live demoCreates plain-language explanations and translation-ready patient communication.
Live demoAn AGx agent is a diligent digital orchestrator — it gathers, monitors, drafts, and flags, so a clinician can spend their judgment on the patient instead of the legwork. It reaches into digital systems around the patient, never the patient. Every clinical decision escalates to a human.
That separation isn't a limitation to apologize for. It's the safety architecture — and it's why the colour on this page means something.
Short answers for clinicians, builders, and patients trying to understand how AGx differs from ordinary medical chatbots.
AGx is a physician-led agentic medicine project where clinical AI agents monitor information, call clinical tools, draft plans, and escalate decisions to a clinician.
A chatbot waits for a prompt. An AGx agent runs a loop around a clinical task: it watches for triggers, uses deterministic tools, drafts the next action, and hands decisions to a clinician.
The page includes 10 live demos spanning the full patient journey — from medication review and pre-visit prep through clinical reasoning, referral preparation, chronic disease monitoring, patient recall, discharge reconciliation, and patient communication.
No. AGx drafts and proposes. Every clinical decision remains human-in-the-loop and must be verified by a licensed clinician.