Security at Rendel: what we keep, where it lives, and what nothing happens without
Last updated 13 October 2026
Rendel keeps conversations and what your actions returned, stored in Frankfurt. Emails, phone numbers and card numbers that users type are masked before the model sees them, nothing changes in your systems without the user's confirmation, and every action is on the record for your team.
What we keep
Conversations, and what your actions returned during them. Not your systems, and not your customers' records: Rendel calls your API as the signed-in user and keeps only what the conversation needed.
Retention is yours to set per app, and you can erase a person yourself from the console.
Masked before the model sees it
Emails, phone numbers and card numbers that users type are masked before the model sees them. IBANs and Turkish ID numbers are masked too.
With Strict PII on, names and addresses are masked as well, and files are refused.
Where your data lives
Stored in Frankfurt, in the EU. The model runs at Anthropic in the United States and is not trained on your traffic.
For Turkey: after an assessment, a Turkey region or your own servers, by separate enterprise agreement.
Nothing changes without a yes
A change, such as a transfer or a card freeze, needs a confirmation card that our server builds from your action, never from the model's wording. The server records the answer with the time, the device and the person.
Every action on the record
For each action: when, who, on which device, the confirmation and the outcome. Owners can export it.
No made-up numbers or links
Rendel cannot hand your users a phone number, a time, an address, a map pin or an image that did not come from your own data.
Your data stays yours
Two layers keep each customer's data apart: every query runs in the app's own tenant context, and the database enforces it again. Both are tested on every change.
Who processes data
The companies that process data for Rendel, and what each one sees.
| Company | What for | Where | What it sees |
|---|---|---|---|
| Anthropic | The model that writes each answer | United States | Masked conversation text, the app's action results and attached files, for each turn. Not used for training. |
| Supabase (on AWS) | Database, sign-in for the console, and file storage | Frankfurt, Germany (AWS eu-central-1) | Everything Rendel stores: conversations, action records, files, console accounts. |
| Vercel | Hosting for the API, the console and the website | Functions in Frankfurt (fra1); global edge network | Requests in transit, including end users' IP addresses. |
| OpenAI | Embeddings for knowledge search | United States | Masked end-user questions and the app's knowledge content, to turn them into search vectors. |
| Resend | Email: sign-in links, invitations, limit alerts, inbound support mail | United States | Console users' email addresses and the emails we send them. |
| Stripe | Billing | United States | The organization's id and billing email, and payment details entered on Stripe's own form. |
| GitHub | Optional: reading a customer's repository during setup, if they connect it | United States | The source code of a repository the customer chose to connect. |
| Optional: "Sign in with Google" for the console | United States | A console user's Google account email and name, when they choose that sign-in. |
What we do not have yet
We hold no certifications, such as SOC 2 or ISO 27001.
There is no penetration test report yet; we commission one when a customer asks.
There is no uptime commitment on current plans. Our target is 99.5%.
The console has no MFA or SSO yet.
No data is stored in Turkey today.
Reporting a vulnerability
Email support@rendel.ai with SECURITY in the subject. We answer within one working day.
Questions
Does the model see my customers' card numbers?
No. Emails, phone numbers and card numbers that users type are masked before the model sees them.
Can the assistant move money without the customer?
No. Every change waits for the customer's confirmation on a card our server builds, and runs with their own token under your API's rules.
Is the data used to train models?
No. Your traffic is not used to train models.