Every action scoped, approved and logged.
Every layer of Uniportal is designed with security at its core. The assistant inherits only the technician's permissions, and nothing that changes data runs without a person's approval.
Admins decide which tools and actions an agent can use.
Integrations you don't use stay locked. Turn each action on or off, scope it to an access group, and change permissions per role or team. Anything not allowed is denied by default.
| Action | Access group | Status |
|---|---|---|
| ConnectWise · update ticket | Service desk | On |
| IT Glue · read documents | Everyone | On |
| Entra ID · reset password | Service desk | On |
| NinjaOne · run script | Senior engineers | On |
| Entra ID · assign admin role | Nobody | Off |
Nothing that changes data runs without a yes.
Every action that modifies data needs explicit confirmation from a person before it runs. That is enforced in code, not just in policy, and every decision is logged to the ticket.
The assistant gets the technician's permissions. Never more.
Assistants inherit the exact permissions of the technician who invokes them, with no escalation. Boundaries apply at four levels.
The AI never sees a credential.
The agent never stores or sees raw credentials or API keys. It asks the Uniportal backend to make a tool call, and the backend checks permissions before running the call with credentials it keeps encrypted.
Start locked down. Expand as trust grows.
The assistant can't act without a technician's approval. Start restricted, and give it more autonomy over time.
Human in the loop by design
The assistant cannot act without technician approval. Start restricted and expand autonomy over time.
SSO and role-based access
Single sign-on, plus access controls by role and group.
Permission inheritance
The assistant takes on the technician's scope, limited by customer, user, device and action type.
Full audit trail
Every technician and AI action is logged with timestamp, actor, client context and outcome.
AI chat oversight
Admins can review AI chats across every technician, for training gaps and quality control.
Full agent reasoning traces
Each session keeps a complete trace of the assistant's reasoning.
Encrypted credential storage
Credentials and API keys are encrypted at rest. The assistant has no direct access to secrets.
Credential isolation
Every tool call goes through a backend proxy that checks authorization first. No credentials are exposed to the model.
Multi-tenant data isolation
Client data is isolated, with no cross-tenant access at the application or database layer.
Running a security review?
We'll take your security questionnaire and due-diligence questions, and walk you through how every action is scoped, approved and logged.