Last updated: September 4, 2026
Your agent talks to your customers, so it ends up holding what they told it. This page says where that data lives, who can reach it, how long we keep it, and — at the end — what we do not have yet. It is written to be usable in a vendor review without a call.
AnroAgents runs on AWS in the European Union: Ireland (eu-west-1) for the application, the databases and uploads, Frankfurt (eu-central-1) for the search index and static assets. The operating company, Anro Technologies Limited, is established in Cyprus, inside the EU.
Two things leave the EU, and we would rather state them plainly than have you find them:
Both are covered by the transfer mechanisms in those providers' data processing terms. The full list of who processes what, and where, is at Sub-processors.
One detail that a vendor review will ask about and that is easy to leave out: the model provider keeps a copy of each request for up to 30 days for abuse monitoring before deleting it. It is not used for training, and it runs on the provider's clock — deleting a conversation here does not shorten it.
Conversations are kept for a fixed period after their last message, then deleted automatically — the transcript, the messages, and everything filed under them. The account owner sets the period in Settings, up to the maximum the plan allows.
| Plan | Maximum retention | Shortest you can set |
|---|---|---|
| Free | 30 days | 7 days |
| Starter | 90 days | 7 days |
| Pro | 180 days | 7 days |
| Business | 365 days | 7 days |
Deletion is enforced by the database itself rather than by a script that might not run: every conversation carries an expiry, and a nightly process keeps that expiry in step with the period you chose — including on conversations already stored, so shortening the period reaches history you already have. Deleted data can persist in point-in-time backups for up to 35 days, after which it is gone.
Settings has both rights built in, for the account owner. Export builds a set of JSON files covering the profile, agents, every conversation with its messages, knowledge-base metadata with links to the source documents, billing, team and referrals — machine-readable, which is what Article 20 portability actually means. Credentials for connected services are deliberately left out: those are secrets we hold for you, not data about you, and a copy in a downloaded file is one more place for them to leak from.
Deleting the account takes your agents offline and out of the public registry immediately and revokes your API tokens. The erasure itself happens seven days later, and you can cancel at any point in those seven days — which is there to protect you from a hijacked session and from a regretted click, not to delay the request. What is erased: agents, conversations and messages, the knowledge base in both storage and the search index, uploaded files, integrations, connected domains, team memberships and your sign-in. What survives: invoices and payment records, for as long as tax law requires.
Between customers. Every record is keyed by the account that owns it, and every read and write is checked against the identity making the request rather than against an identifier the caller supplied. An agent retrieves only its own knowledge base. This is the boundary we test hardest, because it is the one that would matter most.
Inside your workspace. Managers you invite see conversations and agents; they do not see billing, and only the account owner can change the retention period.
On your side. Sign-in runs on AWS Cognito at auth.anroagents.com, with email and password or Google sign-in. We never see or store your password. The API key in the widget snippet is public by design — it identifies the agent to the widget and can do nothing else; it cannot read conversations or change anything, and you can regenerate it. A personal access token is the opposite: it manages agents through the API, so treat it as a credential and revoke it from the dashboard if it leaks.
On our side. Access to production is limited to the people who operate the Service, granted with least privilege, and bound by confidentiality. Credentials for third-party services live in an encrypted managed parameter store, never in source control.
Your conversations and your uploaded documents are used to answer your visitors, and for nothing else. They are not used to train or fine-tune models, ours or our providers', and they are never made available to another customer's agent. The terms we contract under with the model provider state that data submitted through its API is not used to train or improve its models — that is the default for its API, and we have not opted out of it.
An AI agent reads text from strangers, so the honest threat model is that some of that text will try to give it instructions. Two decisions bound what that can achieve:
The same reasoning is why an agent stops at answering in a group chat: everything else it can do assumes it is talking to one identified person.
Send it to support@anroagents.com. We acknowledge within two working days and will tell you what we found and when it was fixed. Please give us a reasonable period to fix an issue before publishing it, and please do not run tests that degrade the service for other customers or touch data that is not yours. We do not currently pay bounties, and we will not pursue anyone who reports in good faith under these terms.
If personal data we process for you is breached, we notify you without undue delay and within 48 hours of becoming aware, with what we know about the scope and the effect — enough for you to meet your own 72-hour obligation to a supervisory authority. The commitment is written into the Data Processing Agreement rather than left as an intention.
A security page that lists only strengths is not useful in a vendor review, so here is the other half:
What we do have, and can show you: signed data processing agreements with each provider that touches your data, available under a confidentiality agreement, and the audit rights in our own DPA. If one of these is a hard requirement for you, tell us before you buy: some are on the roadmap, and it is better to know which of them you are waiting for.