PlatformDelivery BoardAutomation & RunbooksOutcome ReceiptsJust-in-Time ElevationCredential VaultGoverned SessionsDevices & DiscoveryPatch ManagementReporting & ExportsRoles & Multi-Tenancy
Verified AI OperationsThe Operation LoopCommanded AutonomyCompare the operating modelThe Verified Operation Spec
SolutionsFor MSPsFor Enterprise & Internal ITHealthcareLegalFinancial servicesMunicipal & Education
ProofOperation walkthroughSecurity & architectureVerified Operation SpecFive questions for your RMM's AIChangelog
CompanyAboutFounder's noteContact
PricingBuy 1–20 technician licenses onlinePlans — from $499 per monthCustom requirementsFoundation Circle
Log in

Home/Solutions/Healthcare

Healthcare

Keep clinical teams moving. Keep IT work accountable.

A waiting room does not pause for a maintenance window. The work that keeps a practice running has to happen at 09:38 on a Tuesday, on machines that sit next to patient data, with access scoped to the operation. Seraph helps run the fix inside clinic hours and produces a record of who allowed it, what rights it used, and what the machine itself said afterwards.

Outcome receiptOP-000217
operationservice recovery
requested byDana K · front desk, Teams
authorized byMarcus R · 09:40
targetSRV-ACCT-02 · print server
admin accessleased 8 min · expired 09:47
ranrestart print spooler · signed job
the machine saidservice running · queue clear
undodeclared before execution
copy sent toPriya S · practice manager
chain 9f3a…c47e DONE & LOCKED

The pressure

The day is the constraint, not the ticket.

Most environments get to say "we will look at it tonight". A clinic cannot. Check-in stops, the schedule slips, and by 10:15 the delay is measured in patients rather than minutes. So the real question is not how fast a fix is. It is which fixes are allowed to happen while people are in the building — and whether anything records the difference.

A clinic day, and where the access actually satOP-000217
ONE CLINIC DAY, DRAWN TO SCALE 09:38 · THE FRONT-DESK PRINTER STOPS 18:10 · THE STEP THAT NEEDED A RESTART 07:00 MORNING CLINIC · 08:00–12:15 AFTERNOON CLINIC · 13:00–17:30 MAINTENANCE WINDOW THE AMBER SLIVER IS THE ONLY TIME ADMIN RIGHTS EXISTED. IT IS EIGHT MINUTES WIDE, AT THIS SCALE. THE SAME EIGHT MINUTES, EXPANDED · 09:36 → 09:52 ADMIN ACCESS · GRANTED 09:40 · EXPIRED 09:47 ON THE CLOCK 09:38 REPORTED 09:40 AUTHORIZED 09:43 JOB RUNS 09:46 CONFIRMED 09:48 SEALED 09:50 USER TOLD THE FRONT DESK IS DOWN · 12 MINUTES, INSIDE CLINIC HOURS THE DRIVER UPDATE THAT NEEDED A RESTART WAS NOT ATTEMPTED HERE. IT WENT TO 18:10.
A step that would sign a clinician out or restart a machine in use is a precondition on the runbook, not a judgement call made in the moment. When a precondition fails, the operation stops and asks a person rather than deciding for you. The amber sliver is drawn at true scale against the day.

The characteristic ticket

The front-desk printer, at the worst possible time.

This is the illustration we use across this site, because it is the most ordinary thing that happens in a clinic and it exposes every part of the problem at once. Dana works the check-in desk at a multi-site practice. At 09:38 the encounter forms and the wristbands stop printing.

  1. 09:38 — Dana asks in the channel she already uses

    She does not open a portal or look up a phone number. She types it into the Teams channel the practice already runs its IT questions through, and the request becomes a typed outcome instead of a paragraph someone has to interpret.

    Intake from Teams, Slack, or your help desk.

  2. 09:39 — a fix that was already proven gets selected

    The planner picks a certified runbook first. It carries its preconditions, the check that defines success, and the way back — all written down before it is eligible to run. Nothing is improvised on a server that sits behind a clinical workflow.

    See how the loop works.

  3. 09:40 — Marcus authorizes it

    Your technician sees what will run, on which machine, with what rights, and for how long. One click. His name is on the work from that moment, and it stays attached to it.

    Authorization and work land in the same record.

  4. 09:40 — eight minutes of admin rights appear

    Scoped to this operation on this one server, with the reason recorded, and set to expire on the clock rather than when the job happens to finish. No shared admin account was used, because there is not one.

    Expired 09:47 — how leases work.

  5. 09:43 — the job runs, and the model never held the password

    The connection is authenticated before any generated step exists. The planner works from the operation, the target and the outcome to reach; the credential is resolved by the executor and is never delivered to the model.

    The keys stay in the vault.

  6. 09:46 — the server answers, and that is what the record carries

    An independent check runs on SRV-ACCT-02 itself. Service running, queue clear — the machine's answer, not the platform's opinion of its own work. At 09:48 the record seals. At 09:50 Dana is told, in the channel she asked in.

    One reference number resolves to the whole operation.

Scope

How small the access actually was.

A clinic network is unusual in one respect: almost every box on it holds or touches patient data. That is why a standing domain admin account is such an expensive convenience, and why the interesting number is not "how many rights" but "how few, and for how long".

One clinic network, two kinds of accessscope = operation × target
ONE CLINIC NETWORK · WHAT EACH KIND OF ACCESS REACHES WHAT A STANDING DOMAIN ADMIN ACCOUNT REACHES EHR APP SERVER patient records IMAGING STORE studies and reports DOMAIN CONTROLLER every account BACKUP TARGET certified runbooks first FRONT-DESK PCs check-in, scheduling CLINICAL PCs ×14 exam rooms PRINT SERVER SRV-ACCT-02 PHARMACY TERMINAL controlled orders WHAT THE LEASE ALLOWED ONE TARGET: SRV-ACCT-02 ONE OPERATION: restart a service ONE NODE, ONE OPERATION, 8 MINUTES EVERY BOX INSIDE THE DASHED LINE HOLDS OR TOUCHES PATIENT DATA. THE OPERATION NEEDED ONE OF THEM.
Privilege here is a product of two things, not one: the operation and the target. A lease that permits restarting a service on SRV-ACCT-02 does not permit restarting a service anywhere else, and it does not permit anything else on SRV-ACCT-02. It also stops at 09:47 whether or not the work finished.
  • No shared admin account. There is no "the IT password" for a technician to remember, share, or leave behind when they change jobs.
  • The reason is recorded with the grant. Not in a spreadsheet next to it — in the record the work produced.
  • Expiry is on the clock. Rights end at a time, not on a job status, so a stalled operation cannot quietly hold privilege open.
  • A technician can take the wheel mid-operation, under a fresh lease in their own name, in the same operation and the same record.

The person who asks

What your practice manager can answer on her own.

Priya runs the practice. She is the one who sits with the annual security risk analysis, the vendor questionnaires that arrive with every new system, and the cyber-insurance renewal. Today her answer to "who touched that server, and what were they allowed to do" is a phone call to you. It does not have to be.

What Priya gets asked Where the answer is produced What she can hand over
Who touched the system, and when?the authorizationA named person and a timestamp, in the same record as the work
What rights did they hold, and for how long?the leaseEight minutes, scoped to one server, expired 09:47
Did it actually work, or did someone say so?the target's own checkThe server's answer, not ours
What if it had made things worse?the declared undoWritten down before it ran; availability depends on the runbook
Could this record have been edited afterwards?the chainEvery later record would stop agreeing
Can I send this to our security officer?the exportPDF, CSV or JSON, unmetered, under your branding
How we talk about regulation The HIPAA Security Rule asks a covered entity to control who has access to systems holding protected health information, to record and examine activity on them, and to be able to show that it does. Seraph is built for those controls — it does not satisfy them for you, and it is not an attestation. We hold no third-party certification. What we give your programme is evidence it can use: authorization, scope, expiry, an independent confirmation, and a record that resists quiet editing. Read the trust centre for exactly what that means, including where we stand on SOC 2.

Straight about scope

What this is, and what it is not.

Clinical software is not our layer. Seraph works at the machine, service, identity and update layer underneath it — which is where the tickets that stop a clinic actually live.

  • Services, print queues, profiles and drive mappings on the machines a practice runs on.
  • Accounts — directory and local managed accounts, including rotation of the ones you designate.
  • Updates as a governed wave with gates that open on verified results, not on a job being sent. See patching.
  • Governed remote access when a person has to be on the machine, launched under authorization and bound to the operation. See sessions.
  • Not an EHR integration. We do not connect to, read from, or write to your clinical system, and we do not process patient data.
  • Not your risk analysis. Seraph produces evidence for it. It does not perform it, and it is not a substitute for one.
  • Not complete session recording. We are deliberately narrow about what a session record covers today rather than promising replay we do not ship.
  • Not certified against any framework. No badge, no attestation, no third-party audit report to wave at you.

One more line we would rather say out loud. A record proves what the operation did; it does not prove the absence of everything else. A technician in a governed session on a machine can still see what is on that screen. What changes is that the session was asked for, authorized by a named person, scoped, time-boxed, and written down — so the question "who was on that machine on the 14th" has an answer that does not depend on anyone's memory. If your counsel needs a business associate agreement in place before any of this matters, bring that to the demo. We would rather answer it in front of you than in a brochure.

Fifteen minutes, on a live system

Bring the question your last risk assessment could not answer.

We will run an operation end to end, open the record it produces, and execute the declared undo while you watch.