How your data is handled
In transit
Traffic between your browser and Ayulex, and between Ayulex and the services listed below, is sent over HTTPS.
Identifier removal
Before a message is stored or sent anywhere, identifiers with a recognisable format are removed: emails, phone numbers, Aadhaar, ABHA, PAN and SSN numbers, record numbers, and labelled name, date-of-birth and address fields. This is a safety net. A name written in running text will not be caught, so please leave identifiers out.
Language model
Your question and the conversation are processed by language models from one provider, in separate steps: one writes the answer, one compares the cited statements with their sources, and one may rewrite sentences behind a serious safety finding. A question not in English, or with an attachment, is first planned by a model; otherwise the plan is made in code. The provider does not use these requests for training. It deletes them automatically within 30 days, and may keep a request flagged for a violation of its usage policy for up to 2 years.
Saved conversations
If you are signed in, conversations are saved to your account in our database so you can reopen them, for 30 days after the last message in each; then they are deleted automatically. You can delete any conversation sooner; deleting it removes its messages. If you are not signed in, the current conversation is kept only in your browser, for up to a day, and is cleared when you sign out.
Attached reports and images
Images you attach are sent to the language model with your question. The text of an attached PDF report is extracted in your browser, identifiers with a recognisable format are removed from it on our server, and it is then sent with your question; a scanned PDF is sent as images of its first pages. Neither is saved with the conversation. Crop out or black out names and record numbers in images before attaching them.
Dictation
A recording is sent to a speech-to-text service to be transcribed and is not stored by Ayulex. The transcript appears in the message box for you to check before sending.
Operational records
We keep the inputs and results of calculator runs, the specialty a question was classified under, which model answered, how many tokens it used and how long it took, the outcome of the automatic checks on the answer (as counts and categories, linked to any rating you give it), the time and IP address of requests (to enforce daily limits and prevent abuse), and any feedback you submit. These records do not include the text of your question.
Corrections
A correction you suggest after marking an answer wrong is stored with your account until a reviewer checks it against the source you gave. If confirmed, the statement and its source are added to Ayulex's library, without your name or the conversation, and can be shown to the answering model for later questions about the same medicines.
Model training
Ayulex does not train or fine-tune models on your conversations.
What Ayulex is not
We would rather you know the limits than discover them.
- Ayulex has not been audited or certified against HIPAA, SOC 2 or ISO 27001, and does not currently sign Business Associate Agreements. It must not be used to store or transmit identifiable patient information.
- Ayulex is not connected to any patient record system. It supports the reasoning of a qualified professional; it does not replace examination, investigation or clinical judgement.
- Sign-in is by email and password or Google. Two-factor sign-in and verification of professional registration are not yet available; the role you select is self-declared.
Services that process data for Ayulex
| Service | Purpose | What it receives |
|---|---|---|
| Vercel | Hosting | All requests to the site |
| Neon | Database (accounts, sign-in sessions, saved conversations, operational records) | Account details, saved conversations, operational records |
| Language model provider | Language models (planning, answering, checking); no training; requests deleted within 30 days, and flagged requests kept for up to 2 years | Conversation text, attached images and the text of attached reports, after identifier removal |
| Sarvam AI | Speech to text | Dictation audio |
| NCBI PubMed and openFDA | Literature and drug label retrieval | Search terms and drug names |
| ClinicalTrials.gov (US National Library of Medicine) | Registered trials, for a question about trials or new treatments | Search terms and drug names |
| Have I Been Pwned | Refusing a new password that has appeared in a public data breach | The first 5 characters of a one-way hash of the new password, never the password itself |
| US FDA RSS feeds | Daily list of new Drug Safety Communications | Nothing from users: a scheduled request for the FDA's public feeds |
| Upstash | Daily limits and caches | Account id or IP address, with a counter; suggestions written for your account, kept up to 12 hours; answers to general reference questions not about a patient, kept up to 6 hours; search terms and drug names, kept up to a day |
| Razorpay | Payments for the Pro plan, where offered | Your account id, name and email, and the payment details you enter in its window (Ayulex never receives card or UPI details) |
| Cloudflare Turnstile | Bot check on sign-in forms | Browser signals |
| Optional sign-in | Google account identity if you choose it | |
| Email provider | Email confirmation and password reset codes, security notices, and replies to enquiries | Your email address and the message |
See the privacy policy for your rights over this data and how to exercise them.
Institutional enquiries
If your hospital or college has data-governance requirements, a security questionnaire, or needs a larger allowance, tell us what you need and we will reply by email.