Approve network requests

Review and approve blocked agent network requests in the OpenShell TUI.

Approve network requests

NemoClaw runs with a strict-by-default network policy. When the agent attempts to reach a host that is not listed in the sandbox policy, OpenShell intercepts the connection and surfaces it in the terminal UI (TUI) for operator review. You decide whether to allow or block the request.

Prerequisites

• A running NemoClaw sandbox.
• The OpenShell CLI on your PATH.
• tmux if you plan to run the guided walkthrough.

Open the TUI

Start the OpenShell terminal UI from the host machine:

openshell term

The TUI displays:

• Sandbox state and inference provider
• A live feed of network activity from inside the sandbox
• Blocked egress requests that are awaiting your decision

For a remote sandbox, SSH to the instance and launch the TUI there:

ssh my-gpu-box 'cd /home/ubuntu/nemoclaw && . .env && openshell term'

How a blocked request appears

When the agent tries to reach an unlisted host, OpenShell blocks the connection immediately and logs the attempt. The TUI shows the blocked request with the following details:

• Host and port of the destination (for example, registry.pypi.org:443)
• Binary that initiated the request (for example, /usr/bin/python3)
• HTTP method and path, if available

The agent does not proceed until you act on the request.

Approve or deny the request

The TUI presents an approval prompt for each blocked request:

• Select Approve to allow the connection. The endpoint is added to the running policy for the current session.
• Select Deny to keep the endpoint blocked. The request fails and the agent receives a connection error.

Approved endpoints remain in the running policy until the sandbox stops. They are not written back to the baseline policy file. To make an approval permanent, add the endpoint to nemoclaw-blueprint/policies/openclaw-sandbox.yaml and re-run nemoclaw onboard.

Session vs persistent approvals

Approval type	How to apply	Persists after restart?
Session	Approve in the TUI	No — resets when sandbox stops
Dynamic	openshell policy set <policy-file>	No — resets when sandbox stops
Persistent	Edit openclaw-sandbox.yaml, re-run nemoclaw onboard	Yes

For instructions on making permanent policy changes, see Customize the network policy.

Run the guided walkthrough

To observe the full approval flow in a controlled session, run the walkthrough script:

./scripts/walkthrough.sh

This opens a split tmux session with the TUI on the left and the agent on the right. The walkthrough triggers a blocked request, then waits for you to approve it in the TUI.

The walkthrough requires tmux and the NVIDIA_API_KEY environment variable to be set.

Baseline policy

The following endpoint groups are allowed by default and will not trigger approval prompts:

Policy	Endpoints	Allowed binaries
claude_code	api.anthropic.com, statsig.anthropic.com, sentry.io	/usr/local/bin/claude
nvidia	integrate.api.nvidia.com, inference-api.nvidia.com	claude, openclaw
github	github.com, api.github.com	gh, git
clawhub	clawhub.com	openclaw
openclaw_api	openclaw.ai	openclaw
openclaw_docs	docs.openclaw.ai	openclaw
npm_registry	registry.npmjs.org	openclaw, npm
telegram	api.telegram.org	Any binary

All endpoints use TLS on port 443.

Related topics

Customize the network policy

Add endpoints permanently to the baseline policy file.

Network policies reference

Full baseline policy reference including filesystem rules and the operator approval flow.

Monitor sandbox activity

Use the TUI and log commands to trace agent behavior.