Privacy architecture

K Pro is designed with privacy and security at its core to protect your data:

• Personal chat history: Your chat history is securely stored and always associated with your user account and organization

• Access control: Only authenticated users can access their own chat history - no one else can view your conversations

• Secure infrastructure: Our database is hosted on a managed, secure cloud infrastructure with strict access controls and network policies

• Data isolation: Data uploaded to K Pro is only visible to you and members of your organization

Owkin's platform enforces data boundaries through a multi-account architecture, ensuring that each customer's data is isolated from others. Effectively, each instance of K Pro is a single-tenant deployment in an isolated account.

Owkin's platform enforces centralized authorization checks before any dataset/tool access using Role-Based Access Control (RBAC).

Customer data is accessed only through Owkin K services/tools (query-based + APIs/SDK); end users do not get direct DB access.

---

K Pro draws on several categories of knowledge sources to contextualize responses to users:

• Underlying data assets analyzed by K Pro. Held either in secured object storage or in connected data platforms (Snowflake, Databricks, etc.). The data model varies by asset and is preserved as-is from the source system; K Pro accesses these through purpose-built connectors rather than re-modeling the data centrally.

• User information. Stored in a relational database (RDBMS), capturing identity and account attributes such as name, email, and organization.

• Interaction logs. Logged in a relational database (RDBMS). Each entry records the user question, tool calls invoked, latency, timestamp, and supporting metadata required for observability and auditability.

• Evaluation and observability telemetry. Streamed to dedicated observability and evaluation platforms, where logs and traces support quality monitoring, regression testing, and incident investigation.

K Pro does not currently maintain a central knowledge graph / vector store as part of the core data model.

User traces can be accessed programmatically via a backend API to support observability. This API feeds the observability platform.