Clinician control
AI suggestions are designed to require explicit clinician review and acceptance. Nothing is applied automatically to the clinical record without your action.
Security & control
Designed for clinical trust
Security, de-identification, and user control are central — not afterthoughts. Psychiatry.Ink is built to support responsible AI use in psychiatric practice.
De-identification workflows
Evidence and external material can be processed through de-identification workflows designed to reduce identifiable content before AI-assisted analysis.
Regional privacy options
Supports configurable identifier storage and regional privacy settings — designed to align with your practice requirements and data handling preferences.
Audit & transparency
AI-assisted actions can be tracked and reviewed. The workspace is designed to support transparent clinical documentation rather than opaque automation.
Encryption, explained
Two keys, one vault — how patient identifiers are protected
You do not need to be a cryptographer to judge our claims. Here is the whole mechanism, step by step, with a concrete example.
On your device
Anna Example
born 12 Mar 1984
Synthetic example patient
AES-256 vault
Private key — never leaves this device
On our server
What we store
k9Xf3…q2C — 1.2 KB ciphertext
Sealed content key
1 · Encrypt on the device
When you save a patient, name and date of birth are encrypted right in your browser with a fresh content key (AES-GCM-256).
2 · Seal the key itself
That content key is then locked with the public half of a key pair (RSA-2048). The private half — the only thing that can unlock it — is created on your device and never transmitted.
3 · Store only ciphertext
The server receives the sealed result: unreadable bytes plus technical metadata. No names, no birthdates, no keys.
AI without exposure
What happens when you use an AI feature
AI assistance is optional and per-action. When you do trigger it, your text passes a mandatory gate before any model sees it:
Your note
Processed for this one request only — never stored on the server.
De-identification
Names, dates, ID numbers, phone numbers and email addresses are removed or replaced.
Residue check
A second, independent scan. Anything that still looks identifying stops the call.
AI model
Sees only the de-identified text. The reply returns to you as a draft for review.
Blocked
If identifying patterns remain after de-identification, the request stops with an error — the model is never contacted.
What you wrote
Ms Anna Example, born 12 Mar 1984, reports sleeping poorly since May; callback 0176 5550123.
What the model receives
Ms [NAME], born [DATE], reports sleeping poorly since May; callback [PHONE].
Synthetic example — the exact replacements depend on the text.
Zero knowledge in practice
What our server can see — and what it cannot
A fair test of any encryption claim is to ask what the operator could read if they tried. Here is our honest answer.
We can see
- That case a1b2c3 has an encrypted snapshot of 12 KB, last updated at 14:32
- Your account email, plan and credit balance
- Technical AI metadata — tokens used, feature, model — never the text itself
- The public half of your key pair
We cannot see
- Patient names or dates of birth — stored only as ciphertext
- Which real person belongs to which case file
- The contents of scanned documents — they never leave your device
- Your private key or your backup passphrase
Data protection
Built around GDPR and clinical confidentiality
Psychiatry.Ink is designed to support data-protection compliance for psychiatric practice in the UK and EU. You stay in control of patient data and how it is processed.
GDPR-aligned data protection
Personal data is processed under the UK GDPR and EU GDPR on clearly defined lawful bases. When you document patient care you act as the controller and Psychiatry Ink Ltd acts as your processor under a data processing agreement. International transfers, where they occur, rely on Standard Contractual Clauses and supplementary safeguards.
Clinical responsibility stays with you
Psychiatry.Ink supports but does not replace clinician judgement. It does not diagnose patients or make autonomous treatment decisions. AI-assisted suggestions always require clinician review and acceptance, and the service is not intended for emergency or crisis use.