Identify and Scope
Resolve identity, tenant, role and permitted task boundary.
Governance and Security
Keep Control While AI Is Working
Sastra keeps identity, tenant boundaries, policy, privacy, approvals, audit evidence and operational override inside the execution path so authorised people can still constrain or correct AI-enabled work.
Where control matters
Control Before AI Acts
Policies matter most when they can still change what information is accessed, which tools are available, whether approval is required and what is allowed to happen next.
Know Who Is Asking
Establish the user or service identity before exposing context, models, tools or actions.
Keep Tenant Boundaries Intact
Use customer, workspace and user boundaries to control which records and memories may be assembled.
Check Policy Before Action
Decide what may be suggested, stored, routed, approved, executed or blocked.
Keep Evidence After Important Steps
Preserve requests, reasoning inputs, policy checks, approvals, actions and exceptions for review.
Safeguards
More Than One Control
Enterprise AI risk is not solved by one switch. Several controls work together around identity, access, policy, approval, evidence and intervention.
Right Access for Each Role
Give users, groups, roles and service identities only the context, routes, tools and actions required for their responsibilities.
Tenant Isolation
Separate data, memory, retrieval and execution according to customer and workspace ownership.
Policy Enforcement
Control model choice, data handling, retention, tool access, execution rights and escalation requirements.
Human Approval
Keep sensitive, exceptional or irreversible actions subject to authorised review before execution.
Visible History
Preserve who requested work, what evidence was used, which controls applied and what happened next.
Operational Override
Keep interruption, rollback and correction paths available to legitimate operators during execution.
Sensitive data
Keep Data Within Boundaries
Exact controls depend on the customer’s data classification, infrastructure, model policy and regulatory obligations.
Filter What Does Not Need to Leave
Identify sensitive fields and keep unnecessary data out of selected reasoning or tool paths.
De-Identify Where Appropriate
Replace sensitive values before selected processing and restore them only inside an authorised boundary when required.
Control What Is Retained
Define which context may persist, for how long and under whose authority.
Use Customer-Side Guardrails
Apply Edge Guard patterns for sensitive-data handling, route restriction, audit capture and operational interruption.
Use Approved Model Endpoints
Route work only to models and regions permitted by customer policy and deployment design.
Protect Data in Transit and Storage
Use encryption according to the selected platform and customer infrastructure controls.
How control is applied
Separate Authority From Intelligence
The platform should make it possible to inspect and interrupt the path before consequences become irreversible.
01 / 06Scroll to explore
Limit the Context
Retrieve only authorised records, documents, memories and workflow state.
Apply Runtime Policy
Permit only approved reasoning paths, endpoints, APIs and operations.
Request Approval
Escalate actions that exceed the system’s delegated authority.
Execute and Record
Run the allowed action and preserve the operational history.
Review and Correct
Use observability, audit evidence and override controls to investigate or change the trajectory.
Operating reality
Controls Need Operational Proof
Compliance depends on configured controls, operating procedures, evidence and clear customer responsibilities, not merely the presence of a feature label.
Verify Each Environment
Test identity, permissions, tenant separation, retention, model policy and approvals for the implementation.
Keep Human Ownership
Business owners, security teams, compliance functions and authorised approvers retain accountability for consequential decisions.
Use Evidence, Not Assumptions
Architecture, logs and control histories can support assurance and audit processes without implying automatic certification.
Continue exploring
See Controls in Context
Governance becomes concrete when it is attached to identity, tools, deployment boundaries and a real workflow.
MedhaOS
See the control plane around identity, policy, approvals, auditability and execution.
Deployment
Compare cloud, private cloud, hybrid and customer-controlled deployment patterns.
Governance Walkthrough
Follow one guided scenario from identity and policy through approval, execution and audit evidence.
Start with one control boundary