HIPAA Compliant • 100% Local AI • Works Offline

Stop typing.
Start capturing.

Kontext photographs clinical documents with your phone and extracts structured data using AI that runs on your network — not in the cloud.

"I spend more time typing into the EMR than talking to patients."
— Every medical assistant, everywhere
MC
EV
RP
Trusted by eye care practices in private beta

Davis, Robert

DOB: 06/12/1978 • MRN-001234

Pre-Testing Progress 3 of 4
👁️

Autorefractor

OD: -2.25 -0.75 x178

👓

Lensometer

Current Rx captured

💳

Insurance Card

BCBS PPO • XYZ123456

📋

Intake Form

Tap to capture

Synced to OfficeMate
🧠
94% confidence

Manual data entry is killing clinic productivity

Every day, skilled medical professionals waste hours transcribing data that's already written down. It's tedious, error-prone, and takes time away from patients.

⏱️

30-60 minutes per patient

Lost to data entry alone

Intake forms, insurance cards, autorefractor printouts, lensometer readings — each one gets manually typed into the EMR. Field by field. Patient after patient. All day long.

"By the time I finish entering one patient's data, the next one is already waiting. I'm always behind." — Medical Assistant, Private Practice

15% transcription error rate

Industry average for manual entry

Sphere confused with cylinder. Member ID off by one digit. Date of birth transposed. Small errors that cause big problems — rejected insurance claims, incorrect prescriptions, compliance issues.

"We had a patient get the wrong glasses because someone typed -1.75 instead of -1.25. It happens more than we admit." — Office Manager, Optometry Group
💸

$47,000/year per MA

Cost of data entry labor

A full-time MA spending 40% of their day on data entry isn't just inefficient — it's expensive. That's skilled labor doing unskilled work, while patients wait and providers fall behind schedule.

"I didn't go to school to be a typist. But that's what I do most of the day." — Certified Ophthalmic Technician
🔒

HIPAA anxiety with AI tools

Cloud AI = PHI exposure risk

You've heard about ChatGPT and AI that can read documents. But sending patient data to OpenAI or Google? That's a HIPAA violation waiting to happen. So you keep typing.

"We looked at AI solutions but our compliance officer shut it down. Can't risk patient data in the cloud." — Practice Administrator, Multi-location Group

AI that reads documents.
Without leaving your network.

Kontext combines the convenience of phone-based capture with AI that runs entirely on your local hardware. Snap a photo, get structured data, sync to your EMR. No typing. No cloud. No PHI exposure.

Capture with your phone

Use your iPhone to photograph printouts, forms, and cards. One-handed, on the move.

Extract with local AI

Ollama vision models run on your Mac Mini. Data never leaves your network.

Review before syncing

AI extracts, you verify. Confidence scores flag uncertain values for review.

Sync directly to EMR

Data flows into OfficeMate, Compulink, or your EMR. No double entry.

📷

Insurance Card
Front captured

Extracted Data 94%
Carrier Blue Cross
Member ID XYZ123456789
Group GRP-8847
Plan Type PPO

Built for how clinics actually work

Every feature was designed by observing real clinical workflows. No theoretical features — just solutions to problems we've seen hundreds of times.

📱

Phone-First Capture

Your iPhone becomes a scanner. One-handed operation for MAs moving between exam rooms. Works in portrait for thermal printouts, landscape for forms.

🧠

Local AI Processing

Ollama vision models (llama3.2-vision) run on your Mac Mini. 10-second extraction. No API calls. No cloud services. No monthly AI fees.

📴

Works Offline

Capture and queue data even when WiFi drops. Automatic sync when connectivity returns. Perfect for clinics with unreliable internet.

👁️

Human Review

AI suggests, humans approve. Confidence scores highlight uncertain values. Edit inline before sync. You stay in control.

🏥

EMR Integration

Direct sync to OfficeMate SQL Server. Compulink support coming. Data lands in the right fields — no copy-paste required.

👥

Multi-User Sessions

Quick PIN login per staff member. 5-minute auto-lock for HIPAA. Full audit trail of who captured what, when.

Four steps. Seconds, not minutes.

📸

Capture

Point your phone at the document. Tap to capture. Thermal printouts, insurance cards, intake forms — all supported.

🧠

Extract

Local AI reads the image and extracts structured data. Sphere, cylinder, axis. Member ID, group number. All fields mapped automatically.

Review

Check extracted values on screen. Confidence scores flag anything uncertain. Edit inline if needed. Approve when ready.

🔄

Sync

Data flows directly into your EMR. No retyping. No copy-paste. Patient record updated in seconds.

HIPAA compliant by architecture, not policy

Most "HIPAA compliant" AI tools rely on BAAs and policies. Kontext is different — patient data physically cannot leave your network because the AI runs locally.

  • 100% Local AI — Ollama models run on your Mac Mini. No API calls to OpenAI, Google, or any cloud service. Ever.
  • On-Premise Only — iPhone + Mac Mini on your local WiFi. PHI travels across your network, not the internet.
  • Encrypted Storage — AES-256 encryption for all locally stored data. Secure even if a device is lost.
  • Session Security — Auto-lock after 5 minutes. PIN re-authentication. No sessions left open.
  • Audit Trail — Complete log of who accessed what patient data, when, and what actions were taken.
🔒

HIPAA Ready

Designed from day one for healthcare compliance

🏠

On-Premise Deployment

Runs entirely within your clinic's network

🚫

Zero Cloud Storage

PHI never uploaded to external servers

🛡️

BAA Available

Business Associate Agreement for enterprise

Every role benefits differently

Whether you're capturing data, verifying insurance, filling prescriptions, or seeing patients — Kontext fits into your workflow.

👩‍⚕️

For Medical Assistants

Capture data on the move, not chained to a desk

  • One-handed capture — Phone fits in your scrub pocket, ready when you need it
  • Room-to-room mobility — No more running back to the workstation to type
  • Batch patient workflow — Queue multiple captures, review all at once
  • Instant autorefractor entry — Thermal printout to EMR in 10 seconds
"I used to dread the stack of printouts waiting after lunch. Now I capture as I go and they're in the system before the doctor walks in." — Beta user, Mill Creek Family Eye Center
1 Patient arrives → Capture insurance card
2 Pre-testing → Capture autorefractor printout
3 Current glasses → Capture lensometer reading
4 Review all → One-tap sync to EMR
💼

For Front Desk Staff

Faster check-in, fewer claim denials

  • Insurance card capture — Member ID, group, carrier extracted instantly
  • Reduce typos — No more transposed digits causing claim rejections
  • Desktop webcam support — Use your workstation camera, no phone needed
  • Patient intake forms — Demographics, history, medications extracted
"We used to have 3-4 claim denials a week from insurance ID typos. Last month we had zero." — Front desk lead, private practice
73% Faster check-in
94% First-pass accuracy
0 Cloud data exposure
👓

For Opticians & Lens Consultants

Accurate Rx data, every time

  • Full Rx extraction — Sphere, cylinder, axis, ADD, prism, PD
  • Progressive lens support — Seg height, fitting cross, corridor length
  • Frame measurements — A, B, DBL, temple length captured
  • Lab order integration — Data ready for your ordering system (roadmap)
"Transposing the axis from 175 to 157 means a remake. That's $200 and an unhappy patient. This catches those errors before they happen." — Licensed optician, retail chain
Extracted Rx Data
SPHCYLAXISADDPRISM
OD-2.25-0.75178+2.00
OS-1.75-1.00005+2.00
🩺

For Optometrists & Ophthalmologists

Walk into the room with data ready

  • Pre-populated charts — Autorefractor, lensometer, history in EMR before you enter
  • Reduced wait times — MAs finish faster, patients flow smoother
  • Verification, not data entry — Glance and confirm vs. wait and watch typing
  • Better patient interaction — More eye contact, less screen time
"I used to wait 2-3 minutes while my tech finished typing. Now the data's there when I walk in. That's 20 extra minutes with patients per day." — OD, solo practice
Before Kontext
3-5 min wait
With Kontext
30 sec
📊

For Office Managers

Reduce costs, improve retention, stay compliant

  • Labor cost reduction — 40% less time on data entry per patient
  • Staff satisfaction — MAs do clinical work, not typing
  • HIPAA confidence — Local AI means no cloud PHI exposure
  • Audit-ready logs — Complete trail of who accessed what, when
"My best MA was about to quit from burnout. She said it was the constant typing. Two months with Kontext and she's staying." — Practice manager, multi-provider clinic
Quick ROI Snapshot
Time saved per patient ~15 min
Patients per day ×25
Hours saved daily 6.25 hrs
Annual labor savings ~$38,000

See your potential ROI

Enter your clinic's numbers to estimate annual time and cost savings.

Annual Time Saved 1,560 hours That's 39 full work weeks
Annual Cost Savings $34,320 Based on MA labor cost
Daily Time Freed 6 hrs Per clinic day
Error Reduction ~85% Fewer transcription mistakes
* Estimates based on 50% time reduction with Kontext. Actual results vary by workflow. Request a custom ROI analysis →

How it works under the hood

A simple, secure, on-premise architecture that IT teams can deploy with confidence.

📱
iPhone PWA
Camera capture Offline queue Review UI
Local WiFi
TLS 1.3
🖥️
Mac Mini (M2, 16GB)
Ollama AI (llama3.2-vision) FastAPI backend Local processing
SQL/TDS
Encrypted
🏥
EMR Database
OfficeMate SQL Server Compulink (roadmap) Direct table writes
🚫
No Cloud Services

Zero external API calls. No AWS, no Azure, no OpenAI. Everything stays on your network.

📡
Network Requirements

Port 8000 (API), Port 11434 (Ollama). Internal network only — no inbound internet.

Performance

~8-12 second extraction time. Ollama runs continuously for instant availability.

What happens when AI isn't sure?

Kontext is designed for uncertainty. When confidence is low, you're in control.

📄 Faded thermal printout Low contrast
Extraction Result 67% confidence
OD Sphere -2.25 98%
OD Cylinder -0.75 95%
OD Axis 178 ✏️ 67%
⚠️ Low confidence — please verify axis value on original printout
🎯

Per-Field Confidence

Each extracted value has its own confidence score. High-confidence fields can be auto-approved; low-confidence fields require review.

✏️

Inline Editing

Tap any field to edit. Original image displayed side-by-side for easy verification. No switching screens.

🔄

Re-capture Option

If the image is too blurry or faded, one tap to re-capture. Previous attempt saved for comparison.

📊

Audit Trail

Every edit is logged: original AI value, edited value, who edited, when. Complete for compliance.

Ready to stop typing?

Kontext is in private beta with select eye care practices. Schedule a 30-minute call to see a live demo and discuss your workflow.

✓ Free pilot program ✓ White-glove onboarding ✓ No commitment required

Frequently Asked Questions

Everything you need to know about Kontext, from technical requirements to compliance.

Kontext currently supports:

  • Autorefractor printouts — Most thermal printers including Topcon, Nidek, Reichert, Marco
  • Lensometer readings — Including progressive lens data, prism, PD measurements
  • Insurance cards — Front and back, all major carriers
  • Patient intake forms — Demographics, medical history, medications
  • Prescription pads — Handwritten Rx (accuracy varies with legibility)

We're actively expanding support based on beta user feedback. Request specific document types at vinay@getlegaci.com.

Accuracy varies by document type and quality:

  • Autorefractor thermal printouts: 94-98% accuracy on clear prints
  • Insurance cards: 92-96% accuracy
  • Lensometer readings: 91-95% accuracy
  • Handwritten intake forms: 75-85% accuracy (highly dependent on handwriting)

Every extraction includes per-field confidence scores. Low-confidence fields (under 80%) are automatically flagged for human review. The system is designed for human verification — it assists, not replaces, your clinical judgment.

Typical extraction times on a Mac Mini M2 with 16GB RAM:

  • Simple documents (insurance cards): 6-8 seconds
  • Medium complexity (autorefractor): 8-12 seconds
  • Complex documents (multi-page intake): 15-25 seconds

The Ollama service runs continuously, so there's no cold start delay. Extraction begins immediately when you capture an image.

Kontext is designed for unreliable connectivity:

  • Capture works offline — Images are stored locally on your phone
  • Queue persists — Captured images wait in a sync queue
  • Auto-sync on reconnect — When WiFi returns, queued items process automatically
  • No data loss — IndexedDB stores everything locally until sync confirms

Note: AI extraction requires connection to the Mac Mini on your local network. Extraction happens when connectivity is restored, not in the cloud.

Currently supported:

  • OfficeMate (Eyefinity) — Direct SQL Server integration, writes to patient records

On the roadmap:

  • Compulink — In development, Q2 2025
  • RevolutionEHR — Planned
  • Crystal PM — Planned

Don't see your EMR? Contact us — we prioritize based on demand and technical feasibility.

Yes! While the iPhone PWA is optimized for mobile capture, you can also:

  • Use any device with a camera — iPad, Android phone, laptop webcam
  • Access via browser — Works on Chrome, Safari, Edge on any device
  • Document scanner integration — Import from scanner apps (roadmap)

The iPhone is recommended for MAs moving between rooms. Front desk staff often prefer desktop webcam capture for insurance cards.

Yes. For optical data, Kontext extracts:

  • Basic Rx: Sphere, cylinder, axis for OD/OS
  • Progressive data: ADD power, seg height, corridor length
  • Prism: Amount and base direction (BU, BD, BI, BO)
  • Measurements: Monocular PD, binocular PD, OC height
  • Frame data: A, B, DBL, temple length when visible

Complex prism prescriptions with both horizontal and vertical components are supported. The system flags unusual values for verification.

Kontext is built for this scenario:

  • Per-field confidence: Each value shows its confidence score (e.g., "Axis: 178 — 67%")
  • Visual flagging: Low-confidence fields highlighted in yellow
  • Side-by-side view: Original image displayed next to extracted data
  • Inline editing: Tap any field to correct it
  • Re-capture option: Take another photo if the first was blurry
  • Audit trail: Original AI value, edited value, and editor recorded

Nothing syncs to your EMR until you explicitly approve it. You're always in control.

Required:

  • Mac Mini M2 (16GB RAM minimum) — Runs Ollama AI model. ~$599 new, refurbished available.
  • iPhone (iOS 15+) — Any iPhone from 2019 or later. Staff likely already have one.
  • Local WiFi network — iPhone must reach Mac Mini. No internet required after setup.

Optional:

  • iPad — For larger review screen at reception
  • Mac Mini M2 Pro (32GB) — For higher throughput or larger models

Not required:

  • Internet connection (after initial setup)
  • Cloud services subscription
  • Dedicated server room

Internal network only (no inbound internet required):

  • Port 8000 — FastAPI backend (HTTPS)
  • Port 11434 — Ollama AI service
  • Port 3000 — Development/preview server (optional)
  • Port 1433 — SQL Server for EMR sync (outbound to EMR server)

All traffic stays on your local network. The Mac Mini does not need to be exposed to the internet. We recommend placing it on the same VLAN as your EMR server.

Frontend (iPhone PWA):

  • Updates automatically when you open the app (service worker)
  • No App Store approval needed
  • Rollback available if needed

Backend (Mac Mini):

  • Notification when update available
  • One-command update: ./scripts/update.sh
  • Takes ~2 minutes, minimal downtime
  • Automatic backup before update

AI Model:

  • Models updated via Ollama: ollama pull llama3.2-vision:11b
  • Optional — only when new model versions improve accuracy

Built-in resilience:

  • Offline queue: Phone stores captures locally — no data lost
  • Auto-restart: Services configured to restart on failure
  • Health monitoring: Dashboard shows service status

For planned maintenance:

  • Staff can continue capturing to offline queue
  • Extraction resumes when Mac Mini is back online
  • Typical restart time: under 60 seconds

For hardware failure:

  • Captures preserved on phones until new hardware
  • Configuration backup recommended (we can help restore)
  • Optional: standby Mac Mini for high-volume practices

Yes. The Mac Mini M2 can handle:

  • 5-10 concurrent users comfortably for typical workflows
  • AI extractions queue automatically (first-in, first-out)
  • Each user has their own session with PIN authentication

For larger practices (10+ concurrent MAs), we recommend the M2 Pro with 32GB RAM or a second Mac Mini for load distribution.

Connection setup:

  • SQL Server credentials stored locally on Mac Mini (encrypted)
  • Connection uses TDS protocol over port 1433
  • We recommend a dedicated SQL user with write access to specific tables only

Required SQL permissions:

  • INSERT on patient_exam, insurance_info, and related tables
  • SELECT for patient lookup and verification
  • No DELETE or schema modification needed

We provide a setup script and work with your IT team to configure credentials securely during onboarding.

Primary model: llama3.2-vision:11b

  • 11 billion parameter vision-language model
  • Optimized for document understanding and OCR
  • Runs locally via Ollama — no API calls
  • Requires 10-14GB RAM during inference

Fallback model: moondream:latest

  • 1.6 billion parameters — smaller, faster
  • Used for quick previews or lower-spec hardware
  • Slightly lower accuracy but 3x faster

Model selection is automatic based on document complexity and hardware capability.

Yes — by architecture, not just policy.

Unlike cloud AI tools that rely on BAAs and policies, Kontext is designed so PHI physically cannot leave your network:

  • 100% local processing: AI runs on your Mac Mini, not cloud servers
  • No external API calls: Zero data sent to OpenAI, Google, AWS, or any third party
  • On-premise only: All traffic stays on your local WiFi
  • No PHI storage outside your control: Images and data stored on your hardware

We provide a Business Associate Agreement (BAA) for organizations that require one, though technically no PHI is transmitted to us.

At rest:

  • iPhone: IndexedDB encrypted by iOS device encryption
  • Mac Mini: FileVault 2 (AES-256) recommended for disk encryption
  • Sensitive config: Credentials stored in macOS Keychain

In transit:

  • Phone ↔ Mac Mini: TLS 1.3 over local network
  • Mac Mini ↔ EMR: TLS/SSL for SQL Server connection

Note: We never have access to your encryption keys. All encryption is managed locally on your hardware.

Configurable retention policies:

  • Default: Images deleted from phone after successful EMR sync
  • Optional: Retain images on Mac Mini for X days (configurable)
  • Audit logs: Retained for 7 years (HIPAA requirement) — logs contain metadata, not PHI

Manual controls:

  • Admin can purge all images older than N days
  • Individual patient data can be deleted on request
  • Complete data export available for compliance requests

Every action is logged:

  • User authentication: Login, logout, failed attempts, session timeout
  • Data access: Who viewed which patient, when, from which device
  • Captures: Timestamp, user, document type, patient association
  • Edits: Original AI value, edited value, editor, timestamp
  • Syncs: What data synced to EMR, when, success/failure

Log access:

  • Admin dashboard for viewing and searching logs
  • CSV/JSON export for compliance audits
  • Integration with SIEM tools (roadmap)

Current status:

  • SOC 2 Type II: In progress, expected Q3 2025
  • HITRUST: On roadmap for 2026

Why it's less critical for Kontext:

SOC 2 and HITRUST primarily certify cloud service providers' handling of customer data. Since Kontext processes all PHI on YOUR hardware (not ours), the certification burden shifts.

What we do provide:

  • Security architecture documentation
  • Penetration test results (annual)
  • BAA for organizations requiring one
  • Vendor security questionnaire responses

Your data, your control:

  • All data stored on YOUR hardware — we have no copy
  • Export all data in standard formats (JSON, CSV) before offboarding
  • Uninstall scripts cleanly remove Kontext software
  • Audit logs can be archived for compliance retention

Since we never store your PHI, there's nothing for us to delete. Your EMR already has the synced data, which remains unaffected.

Private beta (current): Free for select practices

General availability pricing (planned):

  • One-time setup: Includes hardware configuration, EMR integration, staff training
  • Monthly subscription: Based on practice size (providers/locations)
  • No per-extraction fees: Unlimited captures and extractions
  • No cloud fees: AI runs locally, no API costs passed through

Contact vinay@getlegaci.com for current pricing. We're working with beta users to establish value-based pricing.

Full white-glove onboarding:

  • Hardware setup: We configure your Mac Mini (or help you purchase one)
  • EMR integration: SQL connection setup, field mapping, testing
  • Network configuration: Work with your IT to open necessary ports
  • Staff training: 1-hour session for MAs, front desk, and admin
  • Workflow optimization: Customize for your specific document types
  • Go-live support: On-call support for first 2 weeks

Included in subscription:

  • Email support (response within 24 hours)
  • Software updates and security patches
  • AI model updates for improved accuracy
  • Documentation and video tutorials

Premium support (optional):

  • Phone support during business hours
  • 4-hour response SLA
  • Dedicated account manager
  • Quarterly business reviews

Yes — structured pilot for qualified practices:

  • Duration: 30-60 days
  • Scope: Full system with 1-2 document types
  • Cost: Setup fee only, subscription waived during pilot
  • Success criteria: Defined together (time savings, accuracy, user adoption)
  • Exit clause: No commitment if success criteria not met

We're selective about pilots — we want practices where we can deliver clear value. Contact us to discuss fit.

Options for evaluation:

  • Live demo: 30-minute screen share showing real extraction
  • Test with your documents: Send us sample images (redacted), we show extraction results
  • On-site demo: For practices within driving distance of Seattle
  • Pilot program: Full deployment with success criteria (see above)

Email vinay@getlegaci.com to schedule any of the above.