Shamar Health · Category: Clinical Agentic OS

Every clinic just installed an AI that writes medical records. Nothing checks its work.

Shamar is the governance-native operating system for clinical care. It starts with Guardian — the safety layer that audits every AI-generated note — and grows into the OS that orchestrates specialized agents across the entire care continuum.

Clinician veto on every action · Full audit trail · Human primacy by architecture
27 hrsthe workday U.S. primary care would need to complete all recommended care and admin tasks
43%of U.S. primary care physicians report burnout — highest among 10 high-income countries
97%of patient information in the EHR is never seen by the clinician
2.5 hrs/wkstill spent on after-hours documentation — even after ambient AI scribes

Scribes reduced typing. They did not reduce thinking, checking, coordinating, or liability. That layer was never built — until now.

Signature demo · one encounter, two systems

Watch Shamar run a real visit.

Mr. Johnson, 65 — heart failure and diabetes. Press run. The left pane is the room. The right pane is the OS: agents activating, evidence surfacing, every step logged. You hold the veto at the end — exactly as the clinician does.

SHAMAR://ENCOUNTER-CONSOLE · PT-4471 · ROOM 3

The room — transcript

The OS — agent activity & audit trail

AUDITsystem idle — every agent action is timestamped, attributed, and clinician-reviewable

Notice what Shamar never does: it never tells the clinician what to prescribe. It surfaces the patient's own data, the label, and the guideline — with the full basis one tap away — and the clinician reasons independently. Transparency isn't a feature. It's the architecture.

The wedge → the OS

Nobody buys an operating system on day one. They buy a seatbelt.

Shamar enters where the pain is sharpest — an unaudited AI is writing your legal medical records — and expands, tier by tier, into the operating system care runs on.

START HERE

Stage 1 · The Wedge

Guardian

$99 /provider/mo

  • Audits every AI-generated note before signature — from any scribe you already use
  • Contraindication & interaction flags with full evidentiary basis
  • Coding-gap and denial-risk detection (ICD-10, CPT, HCC)
  • Accountability Ledger: cryptographically timestamped audit trail
Request Guardian pilot

Stage 2 · Expand

Shamar Clinic

$299 /provider/mo

  • Everything in Guardian
  • Documentation Agent — structured notes in under 3 seconds
  • Care Coordination Agent — scheduling, tasking, patient summaries
  • Built for independents, FQHCs & rural systems the enterprise vendors won't serve
Join the waitlist

Stage 3 · The OS

Shamar Enterprise

Custom + revenue-integrity share

  • Full agent workforce across the care continuum
  • EHR demoted to system-of-record — FHIR in, FHIR out
  • Composable Skills Library & partner marketplace
  • Priced against recovered revenue, not added cost
Talk to us
Governance-native architecture

Safety isn't bolted on. It's Layer 1.

Every competitor is retrofitting agents onto a documentation product. Shamar was built in the opposite order — the governance kernel came first.

LAYER 01

Restricted Execution Environment

A sandboxed kernel that constrains what every agent is permitted to do. Permission boundaries per agent, human veto gates on consequential actions, and a tamper-evident log of everything. A hallucinating agent cannot act outside its cage.

LAYER 02

Document-Centric Interaction

Agents collaborate through one shared, live clinical document — not chaotic message-passing. Every edit is attributed. The clinician sees one source of truth and the reasoning behind every change in it.

LAYER 03

Longitudinal Page-Indexed Memory

Fourteen years of chart is not a vector-search problem. Manifest-guided retrieval navigates records by episode of care — benchmarked at 2.2× the precision of metadata-filtered RAG on longitudinal queries.

LAYER 04

Composable Medical Skills Library

Discrete, validated, versioned clinical actions — generate_soap_note, check_drug_interaction, schedule_follow_up, flag_denial_risk — that agents call and institutions can extend. This is where the OS becomes a platform.

The agent workforce

A digital clinical team — with a chief resident in charge.

Twelve specialized agents, orchestrated. Each with a narrow mandate, explicit permissions, and a name for what it actually does.

🎛️

The Orchestrator

chief resident

Activates specialists by context, integrates their output into one coherent note, routes only frontier-model reasoning where it's needed.

🧠

The Thinker

clinical reasoning

Extracts and validates clinical facts in real time — timestamped, confidence-scored, linked to the dialogue that produced them.

📚

The Librarian

evidence

Surfaces guidelines, labels, and local de-identified cohort evidence — always with the full basis inspectable, never as a command.

✍️

The Scribe

documentation

Structured, specialty-tuned notes in under three seconds, grounded in the verified transcript — never in imagination.

💰

The Biller

coding & revenue

ICD-10, CPT, HCC and E/M codes attached with confidence scores; denial risk flagged before the claim leaves the building.

🧭

The Navigator

care coordination

Books follow-ups, tasks nursing, drafts plain-language patient summaries in the patient's own language and reading level.

🛡️

The Guardian

safety & governance

Validates every recommendation against guidelines, flags contraindications, and maintains the medicolegal audit trail.

⚔️

The Adversary

red team · NEW

Runs continuous adversarial simulation against the other agents — failure modes are found in the sandbox, never in the clinic.

📨

The Advocate

prior auth · NEW

Drafts prior-authorization packets from note evidence and predicts denials before submission — EHR-independently.

🔭

The Sentry

panel watch · NEW

Between visits, watches labs, messages, and device data against each care plan — and escalates only on deviation.

👋

The Greeter

intake · NEW

Interviews the patient conversationally before the visit — voice or WhatsApp — and hands the Orchestrator a structured brief.

🎓

The Apprentice

feedback loop · NEW

Every clinician veto is a labeled training signal. Shamar gets better every day it runs — per clinician, per specialty, per site.

Not a better scribe. A different species.

Scribes document what happened. Shamar governs what happens next.

DimensionAmbient AI ScribesShamar — Clinical Agentic OS
Primary functionTranscribeOrchestrate, audit, and act
TimingPost-visitPre-visit → in-room → between visits
Error catchingNone — the note is signed as generatedGuardian audits every note before signature
Reasoning transparencyOpaqueFull evidentiary basis, one tap away
Action executionNoneAutonomous within veto gates you set
MemorySession-onlyLongitudinal, episode-indexed
AccountabilityUndefinedCryptographic ledger of every action & veto
Who it servesEnterprise health systems onlyIndependents, FQHCs, rural, and emerging markets too
Integration · nothing gets ripped out

Shamar works beside your EHR on day one. Reads it in month two. Writes to it when you're ready.

Integration friction is where clinical software goes to die. Shamar is architected so value starts at zero integration — and deepens on your timeline, not ours.

LEVEL 0

Guardian sidecar — zero integration

Guardian audits notes before signature from a side panel beside any EHR and any scribe. No APIs, no vendor approval, no IT committee. Live in days, not quarters.

LEVEL 1

Read — SMART on FHIR

Shamar reads the chart through the standard FHIR APIs certified EHRs already expose — medications, labs, problems, encounters. This powers the pre-visit brief across Epic, Oracle Health, athenahealth, eClinicalWorks and more.

LEVEL 2

Write-back — CDS Hooks + FHIR write

Notes filed, orders drafted, follow-ups proposed — into your EHR, always behind the veto gates you configure. Delivered through each vendor's app program as your organization approves.

LEVEL 3

The OS — where there's no EHR to plug into

In settings without incumbent EHR infrastructure, Shamar doesn't integrate — it is the system of record. The care continuum runs on it from the first encounter.

Trust, engineered

Built for the boardroom question: "When it's wrong, who's responsible?"

📜

The Accountability Ledger

Every recommendation, its basis, the model version, and the clinician's accept / modify / veto decision — cryptographically timestamped. Documentary proof of independent clinical judgment.

Human primacy, measured

We publish veto rates. A healthy override rate is proof the clinician is in command; a falling error rate over time is the outcomes story. Both are yours to inspect.

🔐

Data that stays yours

Customer PHI never trains foundation models — contractually. Zero-retention inference, BAAs at every layer, SOC 2 program underway, de-identification by expert determination.

Regulatory posture: Shamar is architected to FDA's 2026 Non-Device CDS criteria — every clinical suggestion displays its complete evidentiary basis for independent professional review, and the clinician's judgment is never replaced or time-forced. Autonomous functions ship only behind explicit veto gates. Higher-risk capabilities follow the device pathway, in the open, when they're ready.

The two-front strategy

One OS. Two markets the incumbents can't touch.

Front one · United States

The abandoned 70%

Enterprise agentic AI requires an Epic contract and a health-system minimum. Guardian serves everyone they structurally can't: independent practices, FQHCs, rural hospitals, DPC clinics — starting as the safety layer on the scribe they already bought.

  • Rides the incumbents' installed base instead of fighting it
  • Specialty Skill Packs — nephrology first, pulmonary next
  • Clinical validation through peer-reviewed pilot studies

Front two · Nigeria & emerging markets

The leapfrog

Where EHR penetration is minimal, Shamar isn't an add-on — it is the clinical operating system, the way mobile money leapfrogged banks. Deployed through an existing telemedicine network, payer channels, and voice-first intake built for how care actually happens.

  • Every teleconsult becomes an agentic encounter from day one
  • WhatsApp-voice intake for low-bandwidth settings
  • Agentic AI as clinical workforce multiplication at 1:5,000 physician ratios
Pilot program · limited sites

Be the clinic where the category was proven.

We're onboarding a small cohort of pilot sites for Guardian. Pilot partners get founding pricing, direct access to the clinical team, and co-authorship consideration on published outcomes.

CLINICIAN VETO ON EVERY ACTION · FULL AUDIT TRAIL · NO PHI REQUIRED TO APPLY