Strict mode by default
All text masked with layout-preserving glyphs, inputs masked, images blocked. Allowlist a DICOM viewer canvas with a single attribute and it routes through the PHI path.
Strict masking is the default: every character is replaced on the device before upload, images and canvases are blocked unless allowlisted, and AI runs only inside a PHI enclave. Telehealth workstations record for days in monitor mode.
A synthetic intake session in strict mode. The clinician's note and the patient's name never left the browser as text.
All text masked with layout-preserving glyphs, inputs masked, images blocked. Allowlist a DICOM viewer canvas with a single attribute and it routes through the PHI path.
Dedicated processing, dedicated keys, egress allowlists, and AI processing that cannot leave your residency region. A tripwire alerts on any attempt otherwise.
Every replay view, export, live-assist action, AI query, and recording-setting change is logged immutably and exportable to your SIEM. Masking happens at capture, so there is nothing to unmask later.
15-minute idle timeout for HIPAA tenants, passkeys or TOTP required, SSO with SAML or OIDC, SCIM deprovisioning within 60 seconds.
Cryptographic erasure per patient identifier and a one-click export for access requests, both with receipts.
The same Helm chart runs inside your VPC or data center, air-gapped, with your own model endpoint for AI.
A sitter watches six patient feeds for a twelve-hour shift. Recording every frame is expensive and almost all of it is uninteresting. Monitor mode keeps a 60-second pre-roll in memory and uploads only when something happens: a feed drops, a token refresh fails, an alarm fires.
The minute before the feed dropped, the 503 from the streaming service, the WebSocket closing with code 1006, the alarm state on the tile, and the click that recovered it. Idle segments are free.
A telehealth visit that fails is almost never a DOM problem. The far end sent no audio, the clinician's camera was held by another application, the hospital network forced media through a relay, or a shared workstation had no CPU left to encode. Backstory records WebRTC state and quality next to the replay, so each of those is a named finding rather than a phone call that ends in a shrug.
Media frames are never captured. We record metrics and state only, so the visit's audio and video never leave the device. Candidate addresses are hashed and device labels are masked, because a label like “Dr. Rivera's AirPods” is identifying.
Thirteen detectors run on every call, including one-way media, freeze, audio dropout, device-in-use, and relay fallback. Each doubles as a monitor-mode trigger, so a dropped patient feed on a telehealth shift records the minute before it happened. How call observability works →
Is it our app or their network? Which sites force a TURN relay? Did this release regress setup time? Are the complaints coming from one device model that cannot encode at the resolution we ask for? Which clinicians are hitting a camera conflict with their other software?
The BAA, encryption with per-tenant and per-patient keys, the PHI enclave, audit logging, access controls, data residency, breach-notification procedures, and workforce training evidence for your vendor review.
Masking attributes on the elements that matter, canvas allowlists for imaging viewers, roles for your users, retention per project, and the consent posture for patient-facing surfaces. Privacy Preview shows you the result before a single session is recorded.
Free for 1,000 sessions a month. Strict masking, conditional recording, and Signals are all included.
Free for 1,000 sessions a month, no credit card. Or leave a work email and we will reach out about the private beta for healthcare and support teams.
We'll reach out to about the private beta.
We'll email when Backstory opens.