The short version: the secret is encrypted in the user's browser to one technician's key, we only ever store ciphertext, and only that technician's passkey can unlock it. Here's exactly how, and, plainly, what we can and can't see.
The credential is encrypted in the submitter's browser before any network request. We store the ciphertext envelope only, and hold no key that can decrypt it. Reveal happens only in a technician's browser after a passkey step-up, there is no server-side decryption endpoint to abuse, subpoena or misconfigure.
A static invariant we call I-1 is checked by an automated test that walks every response and log and rejects any envelope field, link token, challenge code or recovery code. If secret material could reach the server, the test fails, so the guarantee is enforced in CI, not just asserted in marketing.
A random content key encrypts the fields with AES-256-GCM; that key is sealed to each recipient with HPKE (DHKEM P-256 / HKDF-SHA256 / AES-256-GCM), via the platform Web Crypto API. Keys are non-extractable where the platform allows, and the envelope format is versioned with a published validator. We describe the primitives, we don't say "military-grade".
Each technician has a portable identity key, not a device-bound one, unlocked by a passkey, a 24-word (256-bit) recovery phrase, or a cached device. Recipients are explicit per request, with an optional org recovery key. A new laptop or a leaver never strands a credential; break-glass recovery and offboarding are built and tested.
Every handover page is branded on your domain with named IT contacts. An optional spoken challenge code read out of band, and an optional emailed one-time code, mean a leaked link alone can't complete the flow, and a public link checker lets a recipient confirm provenance before opening.
A handover is created from your ticket, a private note is written back, and it auto-purges when the ticket closes. That removes the orphaned-secret failure mode entirely, it's a control, not a convenience.
To run the service and give you a complete audit trail, we keep the details of each handover: who asked for it, who it was for, when it was opened and which ticket it belongs to. The password itself is encrypted in the user's browser before it reaches us, and sealed to your technician's key. We never see the password, and we never hold a key that opens it, so there is nothing readable on our side for anyone to find. The technical detail is in our docs.

SafeHanded runs in the European Union.
We hold none yet, and won't imply otherwise. SafeHanded is built to produce evidence for your ISO 27001, Cyber Essentials and SOC 2 programmes.
Responsible disclosure is welcome. A security contact and security.txt are published; the data controller is Harman AJ Ltd.
Records are ephemeral by design, SafeHanded is not a password manager.
No self-service reset, no PAM, no session brokering. It's for the cases a reset can't cover.
Subjects interact through links only, there's nothing for them to sign up for.
Start free and run the verifier against your own audit export. That's the whole idea.