Every layer of Uniportal is designed with security at its core. Here's how we keep your data and operations safe.
Granular integration control
Admins define exactly which tools and actions an Agent can access. Unused integrations stay locked, and permissions can be adjusted per role or team at any time.
Every write action needs approval
Every action that modifies data requires explicit human confirmation before execution. This is enforced in code, not just policy. The assistant physically cannot bypass it.
Assistant proposes action
Execution
Blocked until approved
Audit Log
Permission inheritance by design
Assistants inherit the exact permissions of the technician who invokes them. If a technician can't access a client, neither can the assistant. No escalation, no exceptions.
Assistant inherits invoking user's group permissions
Credentials the agent never sees
The agent never stores or sees raw credentials or API keys. Instead, it requests tool calls through the Uniportal backend, which validates permissions and executes the call using securely stored credentials. If the agent can't see the credentials, it can't reveal them.
Assistant never sees or handles raw credentials
The platform
The assistant cannot execute any action without explicit technician approval. Start locked down and expand autonomy as trust develops.
Single sign-on and granular role and group based access controls scope what each technician can do.
The AI assistant inherits the permission scope of the technician driving it, scoped by customer, user, device, and action type.
Every technician and AI action is logged with timestamp, actor, client context, and outcome. Nothing hidden.
Admins can review every AI chat across all technicians, to see where teams get stuck, spot training gaps, and run quality control.
Every session maintains a complete trace of the assistant's reasoning. Unexpected decisions can be fully investigated.
All credentials and API keys are encrypted at rest. The assistant never has direct access to secrets.
All tool calls route through a backend proxy that validates authorization before execution. No credentials exposed to the model.
Client data is strictly isolated. No cross-tenant data access is possible at the application or database layer.
Security
Every action Uniportal takes is scoped, approved, and logged. A human signs off on every write, and the assistant only ever inherits the permissions of the technician using it.