Trust
One institute, two instruments, one standard. This page is the hub; each product publishes its own detailed model.
Windrose Atlas → atlas.windrose.institute/trustWindrose Logbook → logbook.windrose.institute/security
Windrose Atlas - hosted.
The behavioural oracle runs on our own hardware; an external frontier model receives aggregates only, never your raw records. Your inputs are not training data. Deletion is a protocol, not a promise.
Details: atlas.windrose.institute/trust →Windrose Logbook - on-device.
Recordings, transcripts and notes are processed entirely on the clinician's computer; the default pipeline passes an offline-only test. If a cloud model is deliberately enabled, it receives pseudonymised text only.
Details: logbook.windrose.institute/security →Shared commitments
your inputs never train models
no content telemetry
deletion protocols, not promises
security reports answered within 72 h: security@windrose.institute