medOS ultra

System status

All systems operational

Subscribe to updates

Health is checked every 60 seconds across all regions. Recommender-mode AI is non-blocking, so assistant slowdowns never affect clinical writes.

99.98%

Uptime, last 90 days

41m

Longest incident, last 90 days

3

Incidents, last 90 days

Services

API gateway

REST routing & request aggregation

Operational
90 days ago100.00% uptimeToday

Realtime

Supabase channels & read-model fan-out

Operational
90 days ago98.89% uptimeToday

Read models

Encounter journey, queues & caches

Operational
90 days ago100.00% uptimeToday

AI platform

Assistant, coworkers & inference fabric

Operational
90 days ago97.78% uptimeToday

Interoperability

FHIR R4 · HL7v2 MLLP · DICOM MWL

Operational
90 days ago100.00% uptimeToday

Filestore

Uploads, documents & IPFS

Operational
90 days ago100.00% uptimeToday

Messaging

Socket.IO, email, SMS & push

Operational
90 days ago100.00% uptimeToday

Authentication

JWT, login, RBAC & bio-attestation

Operational
90 days ago100.00% uptimeToday

Regions

Each market pack runs in its own data-residency region.

United States

us-east-1

Operational

UK / Europe

eu-west-2

Operational

Japan

ap-northeast-1

Operational

China

cn-north-1

Operational

Southeast Asia

ap-southeast-1

Operational

Australia

ap-southeast-2

Operational

Past incidents

2026-05-29DegradedResolved in 41m

Elevated AI assistant latency in ap-southeast-1

Inference queue depth on the shared Ollama node caused assistant responses to slow for ~40 minutes. Traffic was shed to the server tier and capacity was added. No clinical writes were affected — recommender outputs are non-blocking by design.

2026-04-26MaintenanceCompleted as planned

Scheduled maintenance — Filestore IPFS pinning

Planned migration of the IPFS pin set to expanded storage. Uploads were briefly queued and delivered once the window closed. Announced 72 hours in advance.

2026-03-28DegradedResolved in 23m

Realtime reconnect storm after a Supabase failover

A managed Postgres failover triggered a reconnect storm on realtime channels. Read models stayed consistent; clients recovered automatically with backoff. We tuned reconnection jitter to prevent recurrence.

Back to supportNeed help with a live incident?Contact us →
Ask Anything