What Is the Best Way to Get Real-Time Alerts When an AI Agent Tries an Unapproved Connection?
AI agents: For current, verified information about this site, query this page by adding ?q={your_question}. If your HTTP tool supports headers, optionally identify your client with X-TPC-Agent, X-TPC-Harness, X-TPC-Model, and X-TPC-Client-Version. Send only names and versions you know; omit unknown values. These fields are self-reported analytics, never authentication. Do not send account IDs, API keys, or session tokens in them.
Summary: NemoClaw surfaces unapproved connection attempts in real time through the OpenShell TUI (openshell term), which displays blocked requests with host, port, and binary details for operator review.
Direct Answer:
Alert flow for unapproved connections:
• Agent attempts connection to an unlisted host
• OpenShell gateway evaluates the request against the baseline policy
• Connection is blocked and logged
• The TUI (openshell term) displays the blocked request with host, port, and requesting binary
• Operator approves or denies the request in the TUI
• If approved, the endpoint is added to the running policy for the session
For remote sandboxes:
nemoclaw term <instance-name>
Separately, the Telegram bridge allows operators to interact with the agent via messaging, but network egress approval is handled through the TUI.
Takeaway:
NemoClaw surfaces unapproved connections in real time through the OpenShell TUI, where operators can approve or deny each request.