Company knowledge.
Considered access.

Louvain runs on Fly.io in the European Union (Frankfurt), with its database on Neon in the same region, and the service that holds sign-in tokens for connected sources runs there too. Some providers process data elsewhere, including the model providers that read questions and passages to write answers, mostly in the United States. Every provider that processes your data, and where, is on our Subprocessors page.

Reading sources, answering questions, the Ask assistant and the Today page send the text each task needs to language models through OpenRouter. Groq runs a small model that classifies questions and checks reply sentences, and Cohere reranks candidate documents. Web search goes to Exa, and entity checks to TypeSafe, only if an admin turns them on. Search indexes and embeddings are built on our own infrastructure. We don’t use your content to train models. We are enabling routing that requires zero data retention where a provider supports it, and the Subprocessors page will say when that applies.

Keys and tokens you paste for a source are encrypted with AES-256-GCM before they are stored. Sign-in tokens for sources you connect with OAuth are held by an authorisation broker inside our own hosting environment, whose admin interface is not reachable from the internet. API and AI assistant tokens are stored only as hashes and shown once. If your Google Workspace admin grants Louvain domain-wide delegation for company mailboxes, Louvain reads only the mailboxes you select, using a service account we operate; the delegation your admin grants in Google is broader than that selection, and they can revoke it at any time in the Google Admin console. We don’t offer customer-managed encryption keys today.

In transit, with TLS, and HTTPS is enforced. At rest, by our hosting and database providers, with keys they manage. Credentials for connected sources are also encrypted by the application, as above.

Production access is limited to named Louvain engineers. We look at workspace content only to give support you ask for, to keep the service secure, or where the law requires it. If support needs to sign in as a user, it uses a single-use link that expires within minutes, and the sign-in is recorded in your workspace’s audit log.

When a connected source reports that an item was deleted, Louvain removes it from search, answers, facts and the keyword index. Removing a connection revokes everything it granted at once, and can delete everything it imported. An owner can delete a workspace, which removes its members, sources, knowledge, conversations and connections, its search-index entries and its permission records, and strips every email address from the audit trail it leaves behind. None of these reach database backups or point-in-time recovery, which expire on their normal schedule, or a model provider’s own logs. When a subscription ends we keep its data for 30 days, so you can export it or reactivate, and then delete it. For anything else, email support and we will remove it.

Owner, admin and member roles; who can read each workspace connection; member removal; an audit log of administrative actions such as membership, connection, token and AI assistant changes; and workspace deletion. Directory sync (SCIM) provisions and removes members for a domain you have verified. On the Business plan, your people can sign in with single sign-on (SAML) through Google Workspace, Okta or Microsoft Entra ID, which an admin sets up in Settings, and you can require it for your domains.

Not yet. Louvain does not currently have a SOC 2 report or ISO 27001 certification. We will answer your security questionnaire, and our Data Processing Addendum describes the controls we commit to.

Yes, as part of a security review: talk to us. The documentation (opens in a new tab) explains who can see what in plain words.

Email security@louvain.ai. We acknowledge reports within 48 hours and give an initial assessment within 7 days. You can test against a workspace you own, as our Acceptable Use Policy describes.