Skip to content

Clinical & Life Sciences · Agentic AI · VDI

The clinical workspace, rebuilt around the clinician.

LocumView is an open-source clinical workspace, built in the open. It packages clinical AI into a signed, portable desktop, designed for independent clinical review, that runs on the hardened host your organization already trusts: alongside your existing Windows EHR desktop, or in place of it.

  • Live: RHEL desktop in the browser
  • Live: Keycloak SSO with MFA
  • Built: signed desktop images with SBOMs
  • Planned: the FHIR stewardship agent
locumview · rhel 10 · gnome
The LocumView GNOME desktop in a browser session: the LocumView mark and workspace dots in the top bar, a mountain wallpaper, the LocumView Tour welcome screen open in the middle, and a dock with Files, Calculator, ONLYOFFICE, Firefox, the tour and the app grid.

Live LocumView session, delivered in the browser.

Two ways to deploy

Start alongside your EHR desktop. Replace it when you are ready.

LocumView is OS-agnostic by design and runs on regulated enterprise Linux hosts your team already supports. Adopt it next to what clinicians use today, then make it the whole desktop where the EHR already runs in a browser.
Start hereLowest risk

Alongside

Clinicians keep their standard Windows EHR desktop untouched. LocumView runs as a companion workspace in a browser tab, and its AI agents connect to the EHR only through sanctioned SMART on FHIR APIs, with your organization’s approval. No rip-and-replace, no disruption to existing workflows.

  • Windows EHR desktop stays exactly as it is
  • Companion workspace in a browser tab
  • EHR data only through approved SMART on FHIR APIs, never from the screen
Natural next stepFull desktop

Instead of

Where the EHR already runs in a browser, LocumView becomes the full desktop: one hardened Linux workspace for the EHR, AI tools, and everything else. Retire per-seat Windows desktop licensing and extend the life of thin-client hardware.

  • One workspace for the browser-based EHR, AI tools, and daily apps
  • Fewer per-seat Windows desktop licenses
  • Thin clients and older hardware stay in service longer

In either mode, the AI layer is the same signed, versioned module, so it can be adopted without changing the desktop and updated without revalidating everything around it.

Design goal. Built: the desktop as a bootc image that CI builds, signs, and ships with an SBOM, that installs and provisions itself, that accepts only signed updates, and that rebuilds from a clean clone with no manual steps. Planned: the FHIR stewardship agent, then independent clinical review of it on synthetic cases. Live today: the RHEL desktop in the browser behind Keycloak SSO with MFA.

The core idea

The workspace is the product. The EHR is one governed data source.

Clinicians already work across a dozen windows. LocumView turns that sprawl into one governed workspace, where the EHR is a trusted source of data rather than the place everything has to happen.

LocumView is designed to connect to the EHR through sanctioned, standards-based APIs: FHIR R4 and SMART on FHIR. No screen scraping, no fragile automation. From that data, the workspace will run AI agents, dashboards, and a coordinated clinical environment around it. This layer is planned and not built yet.

EHRMeditech · Epic · Oracle Health
FHIR R4 · SMART on FHIRSanctioned APIs. No screen scraping.
LocumView workspace
  • AI agents
  • Dashboards
  • Secure messaging
  • Files and documents

Platform

Four commitments that shape every decision.

Security, privacy, access, and reproducibility are not add-ons. They are the foundation the rest of the workspace is built on.
Read the platform overview
  • Secure by design

    Linux desktops hardened to the DISA STIG profile and scanned with OpenSCAP, with every exception documented (done on the reference desktop and built into the desktop image), and each HIPAA technical safeguard mapped to its control, with the gaps stated. Compliance evidence is versioned alongside the code.

    DISA STIGOpenSCAPHIPAA mapping
  • AI that stays inside the boundary

    Planned: local models and agents run inside your environment, so no patient data leaves the regulated boundary. Local GPU inference already runs in the build lab; the clinical agents are not built yet.

    Local inferenceNo external callsLogged
  • Delivered anywhere, through the browser

    Desktops reach users through a browser with single sign-on and MFA. No desktop is ever exposed directly to the internet.

    Browser accessSSOMFA
  • Reproducible and auditable

    The goal: every desktop a versioned, rollback-capable image built from code. Built: the desktop image installs and provisions itself with no manual steps. CI builds, signs, and attaches an SBOM to every image, and desktops accept only signed updates. Proven rebuildable from a clean clone with zero manual steps. The live demo still runs on the hand-built reference desktop, which moves to the image next.

    Image-basedVersionedRollback

Clinical AI

A safety net for short-staffed teams.

Lead example (planned)

Antimicrobial stewardship

When a new culture result posts, an agent will retrieve the patient’s data via FHIR, check it against current guidelines and the local antibiogram, and send a concise, evidence-based summary to the clinician’s inbox. Not built yet; this is the first agent on the roadmap.

  • Shows the data, cites the exact guideline version used, and shows its reasoning
  • Drafts, summarizes, and flags. Never signs, orders, or prescribes. The clinician decides.
  • Every action is logged and auditable
  • Built around transparent clinical decision support principles
See how agents work
Stewardship review
Draft · awaiting clinician review

Synthetic patient · Bed 12 · Med/Surg

New result: blood culture

Organism
E. coli
Susceptible
Ceftriaxone, cefazolin
Current therapy
Meropenem 1 g IV q8h
Renal function
CrCl 68 mL/min

Consider de-escalation to a narrower agent based on susceptibilities.

Local stewardship guideline v2026.1, §4.2 · Antibiogram 2025

Illustrative example. Synthetic data.

Planned agents

  • Planned

    Renal and weight-based dosing checks

    Recalculates doses against current renal function and weight, and shows every step of the math.

  • Planned

    Pharmacovigilance intake

    Helps assemble adverse event reports from chart data for pharmacist review.

  • Planned

    Therapy review

    Surfaces duration, duplication, and interaction questions for scheduled medication review.

Governed knowledge

A governed knowledge base, not a black box.

The design: agents reason only over a curated, versioned clinical library. Each source is dated, owned, reviewed, and signed off by a clinician. Updates are tested against the scenario cases and their reference answers before release, and every change can be rolled back. Planned alongside the agents.
  1. Versioned sources

    Guidelines, local antibiograms, and protocols live in a curated library. Every document is dated and versioned.

  2. Scheduled review

    Each source has an owner and a review date. Nothing ages out of date silently.

  3. Clinician sign-off

    A qualified clinician reviews and signs off every update before it reaches an agent.

  4. Regression testing

    Updates are tested against the scenario cases and their reference answers. Any change in output is reviewed before release.

  5. Feedback loop

    Clinician overrides are logged and reviewed, and every update can be rolled back.

Capabilities

Everything in one workspace.

The target toolset for a single desktop, all self-hosted inside your environment. Each item states what exists today.
  • Browser-based EHR access

    Planned: Meditech Expanse, Epic, and Oracle Health via SMART on FHIR.

  • Clinical AI agents

    Planned: launchable apps that draft, summarize, and flag. Clinicians stay in control.

  • Secure messaging and video

    Planned: Matrix and Element, self-hosted inside your environment.

  • Files and documents

    Live: ONLYOFFICE on the desktop. Next: Nextcloud for shared files and real-time editing.

  • Email, calendar, and contacts

    Planned: Evolution, connected to Nextcloud for shared calendars and address books.

  • Data science and research

    Live: Python. Planned: R, Jupyter, and local model serving for approved research work.

  • Centralized dashboards

    Planned: unit and service views built directly from FHIR data.

  • Curated per role

    Each desktop carries only the apps that role needs.

Third-party names are trademarks of their respective owners and are listed for interoperability only.

Workspace

Familiar from day one.

Your organization sets the layout that matches what staff already know: Windows style, Mac style, or standard GNOME. It is managed centrally and the same for everyone, so there is no new desktop to learn, while the applications, security, and data stay the same underneath. Status: built and locked centrally, on the reference desktop and in the desktop image; the Mac style is verified, the Windows and GNOME styles are built but not yet exercised. The images below are concept mockups. A calm, ad-free desktop greets each user, and a guided first-run tour lets them pick a layout. It can be switched anytime.
  • LocumView desktop arranged in a Windows-style layout with a bottom taskbar and start menu (concept mockup).

    Windows-style

    Taskbar, start menu, and window controls on the right.

  • LocumView desktop arranged in a Mac-style layout with a top menu bar and centered dock (concept mockup).

    Mac-style

    Top menu bar, centered dock, and window controls on the left.

  • LocumView desktop in a standard GNOME-style layout with a minimal top bar (concept mockup).

    GNOME

    The standard GNOME desktop: an overview for apps and windows, no dock or taskbar.

Deploy

Deploy anywhere: your infrastructure, your choice.

Your OS is a parameter, not a rewrite. LocumView is being built from code so that the operating system and the infrastructure underneath it become configuration choices. Today it runs on Red Hat Enterprise Linux; the other targets below are in progress or planned.
Why the choice matters

Data sovereignty starts with who controls the stack.

Clinical data is only as sovereign as the software and infrastructure it runs on. Because LocumView builds on open, auditable Linux and deploys wherever you decide, jurisdiction over patient data stays with your organization, not with a vendor.

  • You choose the jurisdiction

    Data residency is decided by where you deploy: your data center, your cloud account, or a sovereign provider.

  • No single-vendor dependency

    Open, enterprise-supported distributions mean no proprietary OS license or foreign platform holds the keys.

  • Auditable end to end

    Open source operating systems and code-defined builds let your security team inspect exactly what runs.

Operating systems

  1. Red HatDemonstrated

    Red Hat Enterprise Linux

    The reference build. The live demo runs here today, hardened to the DISA STIG profile.

  2. openSUSEIn progress

    SUSE Linux Enterprise · openSUSE

    European-rooted enterprise Linux and a natural fit for sovereignty-focused deployments. In progress: infrastructure work has begun and runs largely in parallel with the RHEL build. Not supported or shipped yet; the goal is the same code with equivalent hardening.

  3. DebianPlanned

    Debian

    Community-governed and vendor-neutral. Intended to build from the same definitions; not built yet, and validation will follow customer need.

Infrastructure targets

  • On-premises virtualization

    Demonstrated on KVM/libvirt, where OpenTofu creates each desktop from the desktop image and removes it again (image signing is next). Planned: Proxmox VE as the production target; designed for the hypervisors already in your data center.

  • Kubernetes

    Live: the access layer (Guacamole, Keycloak, and the tunnel) runs on k3s. Planned: desktops on Kubernetes with KubeVirt.

  • Public or sovereign cloud

    Planned: the cloud your governance allows, with the same images and controls.

  • Thin-client ready

    Because desktops are delivered through the browser, endpoints stay simple: any modern browser works, including on phones. Managed thin clients such as IGEL OS devices should work the same way but have not been tested yet.

USFor US health systems

Built for the infrastructure you already run.

Most hospitals already have capable virtualization, identity, and network teams. LocumView is designed to run on that existing footprint, or in your AWS environment, without a new platform to learn. These are design goals; today it runs in the build lab.

  • Designed to run on existing hospital virtualization and storage
  • Planned: deployment to AWS in your own accounts, regions, and network boundary
  • Aims to reduce reliance on per-seat Windows desktop licensing
  • Designed to fit alongside existing SSO and endpoint management
Deployment details

Security architecture

One path in. Nothing else exposed.

Every session follows the same route, live today: the user’s browser, TLS at the Cloudflare edge, an outbound-only tunnel, an authenticated Guacamole gateway, then the desktop. Remote desktop protocols never touch the public internet.
  1. User browserNo client install
  2. TLS reverse proxyTerminates HTTPS
  3. Guacamole gatewayKeycloak SSO + MFA
  4. Hardened desktopSTIG baseline
public HTTPS onlyprivate outbound-only tunnel
Remote desktop protocols stay on the private network. The dashed boundary marks components that are never reachable from the internet directly.
Security and compliance details
Access
Browser only, via Apache Guacamole
Identity
Keycloak SSO (OIDC)
Directory
Red Hat IdM, federated into Keycloak
MFA
TOTP enforced for organizational accounts; guest demo is password-only and isolated; WebAuthn / FIDO2 next
Transport
TLS at the edge, encrypted tunnel inward, WireGuard between cluster nodes, RDP over TLS
Clipboard
Copy-out blocked, clipboard cleared after 5 minutes (live tests pending)
Backups
Identity server backed up nightly, encrypted, off-host
Network
Outbound-only Cloudflare Tunnel, no open ports
RDP / VNC
Never exposed to the internet
Baseline
DISA STIG + OpenSCAP: applied to the reference desktop and at image build time
Disk encryption
LUKS on image-built desktops, unlocked by the vTPM; cluster nodes next
Audit
Sign-ins and connections logged; session recording next
Portrait of Michael Olszewski

Michael Olszewski

PharmD, BCPS, BCCCP

Clinical Pharmacist Specialist

Clinical AI & DevOps Engineer

Charlotte area, NC

AWS Certified Solutions Architect – Associate (SAA-C03)In progress

The founder

Two decades in acute care, now building the systems behind it.

Michael Olszewski is a dual board-certified clinical pharmacist with two decades in critical care and infectious disease pharmacy. For eight years he led an antimicrobial stewardship program. He is also the engineer who builds LocumView: the identity layer, the desktops, the cluster they run on, and the documentation behind every change.

LocumView comes from the gap between those two jobs: the extra logins, the copied numbers, the alerts that arrive too late or too often.

See it running

See it running.

Request access to the live demo or follow the build in public.

Request demo access

Share a few details and we will email guest credentials for the live demo. Questions about pilots, deployment, or partnerships are welcome too.

[email protected]

Please do not include patient information or any protected health information in this form. Sending opens your own email app with the message filled in; nothing is stored by this website.

Have an account? Log in