Data Security at Klemens.AI: How We Protect Your Company's Information

Why Data Security in AI Isn't an Afterthought

Implementing AI in a company raises legitimate concerns. Where do our documents go? Who has access to them? Could one client's data accidentally "leak" into another's responses? These aren't theoretical questions. Companies entrust Klemens.AI with contracts, regulations, internal procedures, and technical documentation—information whose disclosure could have serious consequences. Therefore, security wasn't an "added-on" add-on. It's the foundation on which we built the entire system.

Data isolation: each company in a separate vault

The most important security principle at Klemens.AI is complete data isolation between companies . What does this mean in practice? When company A uploads its documents to Klemens.AI , these documents go to a separate file store and a separate AI index dedicated exclusively to that company. Company B has its own, completely separate store and index. There is no shared "basket" from which the AI could accidentally retrieve other people's data. This isn't filtering like "show only company A's documents." It's physical separation, and each document repository exists in a separate, isolated environment. Even if a coding error were to occur (and we design the system to prevent this), the architecture physically prevents access to the other company's data. This isolation operates on three levels.

Files

Documents from each repository are stored in a separate, encrypted storage.

AI Index

Each repository has its own search engine that "sees" only its own documents.

Database

Queries are always limited to the context of the company in question.

Precise permissions: not everyone needs to see everything

In enterprises, not everyone should have access to the same documents. HR departments don't need technical documentation, and developers don't need to see contracts with contractors. For this reason, Klemens.AI has a two-tiered permission system :

System permissions

They define what a user can do in the system, such as creating new repositories, managing access, or viewing the activity log.

Permissions to individual resources

They allow for precise control over who can access a specific repository or knowledge base, and at what level. Access levels create a clear hierarchy:

• Preview - the user sees that the resource exists and can view its contents.

• Preview with comments - can also add notes and comments.

• Execution - can ask AI questions and use analytics.

• Full access - can manage the resource, add documents, change settings.

Permissions can be granted to both individual users and groups . The resource owner always has full access, requiring no additional configuration.

Secure file access

Documents stored in Klemens.AI are never accessed via a permanent, public link. Instead, we use signed, time-limited URLs . How does it work?

When a user wants to download a document, the system generates a unique link that:

• It is only valid for a few minutes.

• Only works for this specific user.

• It expires automatically and cannot be saved and used later.

• Contains a cryptographic signature confirming authorization.

Even if someone intercepts such a link (e.g. from browser history), it becomes useless after a few minutes.

Authentication with active verification

Every request to the system requires a valid authentication token. But in Klemens.AI, it's not enough that the token is properly signed; the system constantly checks for revocation . What does this mean? If an administrator locks a user's account or logs them out of all sessions, the next request from that user will be rejected, even if the token is technically still valid. In systems that only check the token's signature, a locked-out user could use the system until the session expires (sometimes for hours).

Full activity log

Every significant operation in the system is recorded in the activity log (audit log). It's not just about recording "who logged in"; we also record:

• Who performed the operation - specific user.

• What did he do - create, edit, delete.

• On what - which document, repository, user.

• When exactly did this happen?

• What has changed - record of the condition before and after the operation.

Sensitive data (like passwords) is automatically masked and ******** appears in the log instead of the actual value.

The activity log is accessible only to authorized administrators and allows for a complete reconstruction of what happened in the system. This is important from both a security and compliance perspective.

European infrastructure

Klemens.AI runs on Google Cloud Platform's European data centers . Documents, AI indexes, and databases are all located in the EU (Western and Central Europe). This isn't a purely technical issue. For many companies, especially those subject to European regulations, data localization is a legal requirement. Klemens.AI meets this requirement by design, not as an optional extra.

Safe removal

When a user "deletes" a document, Klemens.AI does not immediately erase it from the database. We use a soft deletion mechanism : the document is marked as deleted and is no longer visible and accessible to AI, but it physically remains in the system for a specified period of time. Why? Because accidentally deleting an important document is a real risk. Soft deletion allows for data recovery in the event of a mistake. At the same time, the "deleted" document is immediately excluded from the AI index and Klemens.AI will no longer use it in responses.

Security is a process, not a product

No system is "100% secure." This is a statement widely accepted in the IT industry. However, there is a huge difference between a system where security is a fundamental part of the architecture and one where it is an add-on. At Klemens.AI, data isolation, precise permissions, encrypted file access, active session verification, and a full activity log are not "premium" features. They are standard components of every installation, because we believe that corporate data security should not be optional.