Data boundary
Core operation is designed to run inside the customer environment so documents and outputs do not need to move to a public AI service.
Security overview
Private AI architecture buyers can inspect.
ToguAI is designed for sensitive environments where documents, embeddings, model runtime, answers, and logs must remain under customer control.
Identity and permissions
Retrieval is shaped by user identity, roles, attributes, and document permissions before answer generation.
Retrieval and citations
Answers are grounded in retrieved sources and include citations so teams can verify important claims.
Prompt-injection handling
Untrusted instructions found inside documents are treated separately from system policy and permission decisions.
Audit logging
Questions, cited sources, refusals, and policy decisions can be logged locally for review.
Retention
Retention behavior is designed to be predictable and aligned with customer governance requirements.
Software and model updates
Update paths should be reviewed during evaluation so restricted and air-gapped environments can plan controlled changes.
Vulnerability reporting
Security reports can be submitted through the responsible disclosure process.
Evaluation material
Security evidence for private review.
A private walkthrough can include deeper material for technical, legal, and security stakeholders.
- Architecture overview
- Permission model summary
- Threat model summary
- Retention and audit logging explanation