Security

Built so the agent cannot do what you did not allow.

Security in Delvra is not a badge. It is how every request is handled: who it belongs to, what may be read, what may change and who approved it.

This page describes how Delvra is built for the pilot program. Ask us for the current test results of each control.

Organization isolation

Every record, task, file and log carries your organization’s ID, checked on the server against your membership. An ID in a URL grants nothing.

The Action Gate

Every tool call passes one gate that checks scope, role, budget, state and approval. There is no path around it.

Secrets stay secret

Credentials are encrypted, never shown again after saving, never placed in a prompt or a log. You enter them only in the authenticated portal.

Untrusted input

Messages, documents and system responses are treated as data, not instructions. Permissions are enforced by the gate, not by the model’s good behavior.

No open network access

The model never browses or calls arbitrary URLs. Each connector reaches only the hosts approved for it. Network-level blocking of private addresses and redirects is being completed before general availability.

People in control

Approvals are tied to the exact input and expire. Kill switches stop actions per agent, connection or account, checked right before execution.

What we do not claim

Delvra does not hold SOC 2 or ISO 27001 certification at this time, and we will not show badges we have not earned. Ask us for our current security documentation and data processing terms.

Request security information