Information security

Aevy protects your data with private networking, strict access controls, encryption, and regional data handling. This page covers where your data is stored, who can reach it, what we do and don’t do with it, and what happens when the contract ends.

This page and your agreement

Where it conflicts with an agreement we’ve signed with you, including your Subscription Agreement or its DPA, that agreement controls.

Data storage and regions

Sensitive data stores aren’t directly exposed to the public internet, while Cloudflare, TLS, and AWS security controls protect the limited public-facing parts of the service. Aevy segregates each organization’s data from every other organization’s. Unless otherwise agreed, we store and process European data in Europe, meaning the EEA and Switzerland, and not outside it. Our technical and organizational measures (TOMs) name the exact regions, and we send them on request.

We encrypt storage with AES-256, both the database and the files, and encrypt traffic in transit with TLS. Processing stays in region as well, model inference included. Models running on European deployments read a European organization’s documents.

Separation between organizations

Aevy draws separation in the infrastructure, not in a query. Your documents and everything extracted from them are stored in a database and a file store dedicated to your organization. Identity, workspace metadata, and roles are the exception. They’re stored in central core and auth databases, because they’re what authenticates you and lets you move around the product.

Access control

Access to Aevy runs through single sign-on (SAML 2.0 or Microsoft Entra ID), with a Keycloak service issuing the tokens the product checks on every request. Multi-factor authentication, conditional access, and password policy stay in your identity provider. Aevy doesn’t duplicate them here. We store no passwords for your users on our side. Accounts are individual, and there are no shared logins.

1. Permissions follow least privilege. Aevy ties roles to organization membership, so a person sees the assets their role covers and no more.
2. SPV and asset-scoped permissions ship by default. Document-level permissions and other finer controls are available on request.
3. Aevy logs authentication and admin actions in Entra and Keycloak. It logs changes to the information itself separately, in the activity log you can read in the product.

Access at your source

Documents enter Aevy through a read-only sync from the SharePoint sites your administrator grants. It runs as an application with no signed-in user behind it, as Microsoft recommends for background services. The permission model is what bounds it: Sites.Selected conveys access to nothing on its own, and your admin then authorizes one site at a time, read-only. Removing a site’s permission removes Aevy’s access to it. Aevy never writes, moves, renames, or deletes anything at your source, and there is no upload step that could bypass the grant. Connecting SharePoint covers the consent and the per-site grants. See how your data gets in for what happens to a file once it’s synced.

AI, model training, and product improvement

Your data isn’t used by our model providers as training data. We engage each under a data processing agreement that prohibits it, so what they process on your behalf doesn’t feed back into their models.

Aevy’s own models are different. When we train or fine-tune a model ourselves, it runs on anonymized, aggregated data that can’t be traced back to your organization, never your documents.

Improving the product is separate. Your Subscription Agreement grants Aevy a royalty-free right to use feedback and what we learn from your data to improve the product. That covers what breaks, which features get used, and which features we should add. Your data itself stays yours and stays proprietary. We aggregate what we build from that: it describes patterns across many assets, never anything traceable to yours.

Suggestions are drafts you accept or dismiss, and Squirrel answers with the document and page it read.

Sub-processors and certifications

Aevy publishes every sub-processor it uses, what each processes, and its compliance standards, including GDPR, SOC 2 Type 2, and ISO 27001, at aevy.io/trust/subprocessors. A data processing agreement covers each one. That page is the single source for the list. When we add or replace a sub-processor, we update it there and tell you before the change takes effect, in time for you to object.

Aevy follows best practices according to ISO 27001, SOC 2 Type 2, and NIS2, and is working to get certified under these. The TOMs document the underlying controls in detail.

Data retention and deletion

On written request, or at the end of the term, we return or destroy all of your data. That covers the source files, the extracted metadata, and the embeddings and indexes built from them. Automated backups run daily, and we keep them for 30 days in the same region. A deletion clears the live data, and the backups still holding it rotate out inside that window. We keep audit logs on their own schedule, with the exact periods in the TOMs. The Subscription Agreement covers all of this, and we certify return or destruction on request.

Logs and lineage

Aevy keeps a record of what it does with your information, and you can read it in the product rather than asking us for it. The activity log on a record shows who changed what and when, and it records Aevy’s own actions there too, like anyone else’s. Lineage ties every extracted fact back to the line of the document it came from.

Breach notification

Aevy notifies you without undue delay, and no later than 72 hours after becoming aware of a personal-data breach affecting your data, under the DPA and the Service Agreement. The notice states the nature of the breach, the likely consequences, and the mitigations already taken.

Security contact

security@aevy.io