Slate.MD · A clean slate for the reading room

See how a report goes out.

Dictate the way you read: you never speak a comma or a period, punctuation and capitalization land for you. Each finding routes itself to the right organ section as you speak, the engine deciding the section from what you said, not from where the cursor sits. Findings supersede the normal they contradict, and the impression drafts itself from your findings and the patient's prior reports. Call it thirty seconds a report; across a sixty-study shift that is roughly half an hour of your day back, on the reporting side alone. A scripted replay of the shipping workflow, no video.

Start Case Speed ▶ auto-plays when in view

Illustrative replay of the shipping workflow; no audio runs on this page. Grey italic text is the live transcript. Slate decides the organ only after hearing the finding (watch it light up), keeps a continued thought in the same organ, undoes a correction cleanly, and flags a danger pair (medial vs lateral) amber until you click to verify.

The demo shows the transcript half. The same engine drafts the impression from your findings and the patient's prior reports, streaming as it writes: an ungrounded number kills the draft mid-sentence, and a sentence that can't be proven faithful stays in your verbatim words.

Everything around the read, compressed and automated.

Legacy PACS make you glue five products together. Slate.MD is the worklist, the viewer orchestration, the dictation, the structured capture, and the report, with AI working the whole path: findings route themselves as you speak, the impression drafts itself from your findings and priors, and the chart is read for you, never asserting what you didn't say. One system, one browser tab.

Shipping

STAT-first worklist

Deterministic ordering: STAT, then TAT-breached, then oldest. Turnaround targets and breach states are visible per row, tuned per site.

Shipping

Structured measurement modules

Hip/FAI radial reformats with manual alpha, center-edge, and version capture flowing straight into the report. More subspecialty modules follow the same pattern.

Shipping

A viewer built like an instrument

Every monitor filled, every scroll instant, backed by a workstation-memory cache with honest fidelity tags. One keystroke to a 2×2 MPR with draggable oblique crosshairs; slab thickness and spacing live on the function row, W/L presets on F1–F8. It feels tuned because it is: nightly, on a multi-monitor rig, by the radiologist who reads on it.

Shipping

Dictation, on the appliance

Browser mic to on-site speech recognition: instant partials while you talk, a rescoring pass behind them, audio that never leaves the building. Findings route to the right organ as you speak, punctuation and the impression are drafted, and it learns how you say words. Voice runs navigation and signing: say “sign report” to finalize, with an explicit confirm before any instant release. Live now, on real dictation.

Shipping

The best surface to report findings

Dictate findings in whatever order you read; each utterance lands in the right organ section, and the report builds in your own template language. The impression drafts from your findings and the patient's prior reports (yours and a colleague's), while a number grounded in none of them aborts the draft mid-sentence. If a restyled sentence cannot be proven to say exactly what you said, your verbatim words stand. Radiologists lose ~80 minutes a shift wrestling dictation into a report; this workflow takes them back.

Shipping

A vision-language model, working for the radiologist

It reads the prior studies, the requisition, and the chart, and hands you what matters: active problems, meds, recent imaging, labs, each cited back to the source record. We don't know anyone else pointing a VLM at the radiologist's own workflow like this.

Shipping

The clinical story, beside the pixels

A PHI-aware context pane surfaces the patient's story where you read it. A comparable prior's stable findings (even across modality: a CT cyst follows you into the MR) arrive as one-click chips in your own phrasing: "left renal cyst measuring 1.3 cm, unchanged / increased / decreased." A complex or indeterminate lesion on the prior blocks sign until you've measured it today; a simple cyst never holds up your signature. Named interval procedures ("interval ACDF at C5–C6") drop into the right section, attributed, one click to dismiss; a vague "s/p ACDF" stays in the pane rather than guessing. Nothing is asserted for you.

Shipping

Seamless AI-module integration

Slate is the surface best-in-class imaging AI plugs into, not another tool to bolt on. A CT head with a hemorrhage can be at the top of your worklist the moment it lands: the sickest cases surfacing first, results landing where you already read.

Shipping

Mammography workflow

Four-view hanging by view and laterality, BI-RADS capture with density language, and the MQSA audit statistics your physicist asks for. Diagnostic rendering rides a cleared mammo viewer.

Shipping

Pathology review

Whole-slide viewing for tumor boards and consults: gigapixel pyramids, micron-true scale bars, the same appliance. Review use; primary-diagnosis pathology has its own clearance path and we say so.

In development

Cardiac suite

Coronary calcium scoring, vessel centerlines, and LV volumetrics, running today behind an investigational banner while we walk the regulatory path properly. Never in the clinical read.

In development

Automatic segmentation, automatic prior linking

The phases we're building toward: every organ and structure delineated the moment a study lands, and every prior registered to today's case, the same anatomic level under your cursor across years of studies, the lesion you measured last time found and offered, never asserted. Device-track work, behind an investigational banner until it earns its clearance.

Security your CISO can verify, not just believe.

On-prem appliance. PHI never leaves your network by default. The subprocessor list is nearly empty on purpose. And the audit log can prove its own integrity.

$ GET /audit/verify { "valid": true, "checked": 14,203, "brokenAtSeq": null } # Every PHI touch is hash-chained. Edit one event, even one character, # and verification fails at the exact sequence number. Ask your current # PACS vendor to run this query.

Least privilege that means it

Admins cannot open studies. Emergency access exists: break-glass with a written reason, a hard expiry, and loud auditing.

Encrypted, envelope-deep

AES-256-GCM envelope encryption for sensitive payloads on top of full-disk encryption, with key rotation that doesn't re-encrypt your archive.

HIPAA-mapped, in writing

Every control mapped to its Security Rule citation in our architecture whitepaper, with implemented vs. planned labeled honestly. Request the whitepaper.

Integration is where PACS projects die. We planned for that.

Send us twenty sample HL7 messages before the kickoff call. Our conformance linter returns a gap report (which fields map, which priority codes need translation, what your Z-segments carry) so the interface meeting is a confirmation, not archaeology.

HL7 v2, tolerantly

Orders in, results out, with per-site field mapping as configuration, not custom code. Your RIS puts the accession somewhere unusual? That's a setting.

Epic & EHR launch

SMART on FHIR launch from the chart with patient context, plus FHIR reads for the clinical summary. Cerner and Meditech follow the same standard path.

Every kind of hardware

The client probes each workstation and picks its rendering strategy (full client volumes on a workstation, appliance-side rendering for thin clients) with the reasons shown on the diagnostics page.

One rate, per study. Fast quotes. No hostage negotiations.

Incumbent PACS pricing is a ransom note with a renewal date: seat license, modules, interface fees, a maintenance line that grows every year. Ours is one number against the thing you actually do, a study read, and right now it is design-partner pricing, locked on one call. No rates printed yet, by design: the first three groups set them with us and keep theirs for two years past the pilot.

Teleradiology groups

5–50 radiologists, distributed reading

design-partner rate · per study read

Full platform + appliance
Lightweight reader install, minutes to first read
A monthly ceiling per radiologist, so your heaviest reader never becomes a penalty
Founder-direct support line

Imaging centers

Single or multi-site outpatient

design-partner rate · per study, site baseline

Appliance on your hardware or ours
RIS/HL7 interface included
Two-week standard install

Hospitals

Community and critical-access hospitals, directly

same per-study rate, no health-system minimums

Same appliance, hospital SSO
Epic launch + FHIR context
Come directly, or through the terad group that reads for you

Get the design-partner rate

A clean slate
for the reading room.

See how a report goes out.

Everything around the read, compressed and automated.

STAT-first worklist

Structured measurement modules

A viewer built like an instrument

Dictation, on the appliance

The best surface to report findings

A vision-language model, working for the radiologist

The clinical story, beside the pixels

Seamless AI-module integration

Mammography workflow

Pathology review

Cardiac suite

Automatic segmentation, automatic prior linking

Security your CISO can verify, not just believe.

Least privilege that means it

Encrypted, envelope-deep

HIPAA-mapped, in writing

Integration is where PACS projects die. We planned for that.

HL7 v2, tolerantly

Epic & EHR launch

Every kind of hardware

One rate, per study. Fast quotes. No hostage negotiations.

Teleradiology groups

Imaging centers

Hospitals

We're taking three design partners.