Confidential AI governance for data and conversations
Bring your files, analytics and AI conversations into one governed environment on infrastructure you control. Sovrata classifies and secures your data, applies your access rules and sends governed chats to models approved to handle their content.
- hr-medical.xlsx Special category
- customer-master.csv Personal data
- q3-board-pack.pptx Confidential
- brand-guidelines.docx Public
Classify the data. Choose the right model.
Copy your files into Sovrata and classify them against your organisation’s labels. Administrators enable automatic classification, configure its models and move libraries from monitor to enforce mode. Processing runs after content arrives; configured folder and library labels provide the initial classification. Cloud classification assistance is opt-in.
With content routing enabled, a governed chat keeps the highest established security level of the data it has seen. A later harmless question does not lower it. When classification is uncertain, routing uses the person’s highest authorised level. Your organisation decides which models may handle each level.
-
File and search labels
Document search vectors indexed by Sovrata carry their source file’s security classification, including inherited folder or library labels.
-
Approved model routing
You decide which models may handle each security level. Sovrata routes governed chat requests to an approved destination.
-
Connected evidence
Conversation document records and inference audit events connect the sources, person, model and policy decision for supported workflows.
Simple to install. Familiar to use.
Install Sovrata on your servers or in your own cloud account. Copy in your shared-drive files, configure identity and policies, then switch to the governed share. People keep a familiar file-and-folder workflow, with new endpoints and credentials where required.
Connect supported S3 and Apache Iceberg tools for large-scale analytics. Governed reads apply user permissions and classification to tables, columns and data files through the same security framework and shared audit infrastructure. The chosen engine and configuration determine the controls.
Supported NVIDIA Confidential Computing deployments can keep inference data private from cloud infrastructure administrators. The model, connection and processing must stay within the configured trusted environment. Review the confidential inference requirements.
See the source behind each conversation
Review the files and exact versions served to a governed conversation, alongside the person, model and policy decision.
Conversation records show the first and latest delivery of each document. Separate inference audit events record request time, model, security level and the policy decision for successful requests on supported paths. Together they help your security team reconstruct the data and inference history.
Use this evidence to investigate activity and assess how your controls were applied. Review the record coverage and delivery guarantees for the workflow and deployment you intend to use. Explore the current product status and evidence.
