ModelFoundry model hub
All systems operational Sign in

All systems operational

No incidents recorded in the current reporting window. Checks run every 60 seconds from three regions.

Components

What the status page covers

Catalog mirror
Static pages, stylesheets and model cards at the edge
Operational
Manifest API
Read-only catalog index, manifests and evaluation summaries
Operational
Weight storage
Key-gated transfer endpoint; anonymous requests are refused by design
Operational
Evaluation runner
Harness executions for keys with the eval scope
Operational
Documentation build
Rendered from the same repository as the client and runtime
Operational
Key service
Issue, scope, rotate and revoke workspace keys
Operational

Response time, current window

Endpointp50p95p99
GET /18 ms42 ms71 ms
GET /v1/models34 ms78 ms112 ms
GET /v1/models/{id}29 ms66 ms98 ms
POST /v1/pulls (keyed)61 ms140 ms210 ms
POST /v1/pulls (anonymous)2 ms4 ms6 ms

Anonymous pulls are rejected at the edge, which is why they are faster and why they reveal nothing about the catalog.

Availability, rolling windows

WindowCatalogTransfersAPI
Last 90 days100%99.99%99.98%
Last 30 days100%100%100%
Last 7 days100%100%100%
Windows are rolling, not calendar months. A month boundary should never change a number for cosmetic reasons.
Incident history

Nothing to report

Incidents are listed by release line rather than by day, and a quiet page means a quiet fleet.

No open incidents

clear
  • Resolved during the 1.7.1 line: a scheduler edge case that dropped prefix-cache entries after preemption. Rolled into the current runtime long before it affected the mirror.
  • Resolved during the 1.6 line: a storage backend that served a stale manifest to one region for a single reporting interval. Digest checks caught it before any transfer was issued.
  • Planned: transfer endpoints will be drained and re-issued as part of the next storage line. Announced in the changelog before it happens.

Status data is generated from the same probes that guard the mirror and refreshed once per minute. For incident discussion, use the community channels rather than this page.