REFERENCE

Security overview

Review AI NetShell's desktop security boundaries, credential handling, AI data routes, and change safeguards.

Verified against AI NetShell 0.4.11

Purpose#

Understand the security boundaries of the AI NetShell desktop app before connecting devices, adding credentials, using AI, or approving a change.

This page is the product security overview. For legal terms and current data-practice disclosures, see the Privacy Policy and Terms of Service.

Before You Begin#

• Install AI NetShell only on a Windows account and computer you trust.
• Use least-privilege device accounts and provider credentials.
• Connect only to systems you are authorized to administer.
• Obtain SSH host-key fingerprints through a trusted channel.
• Decide whether Included AI or Your own AI fits your organization's data requirements.

Steps#

1. Understand the desktop trust boundary#

The desktop host, not the AI model, owns:

• Device connections and session state
• Credential lookup
• SSH host-key checks
• Command-policy decisions
• Read and change approval state
• Change snapshots, verification, rollback, and local audit records

Lina uses the selected AI model. The model receives context for the requested task and can propose reads or changes. It does not receive a general-purpose SSH connection, a credential-retrieval function, or the ability to approve its own configuration change.

Commands you type directly into a terminal remain direct administrative actions. The AI change protections do not make a human-operated terminal read-only. The same is true of file transfers you start over SFTP, SCP, or TFTP.

AI NetShell account sign-in opens the system browser. The account page asks you to Approve this device, and the app never asks for the account password. The account portal lists approved-device details and provides Revoke; the app itself does not display its device identifier. If several approvals could match an old PC, do not guess which one to revoke. On a PC you still control, use Settings > Account > Sign out to disconnect the app without deleting local devices, saved credentials, images, or history.

2. Understand credential handling#

In the production Windows build, AI NetShell stores enrolled secret values through Windows Credential Manager. This includes saved device passwords or key passphrases, enable passwords, Meraki API keys, provider keys used by the desktop app, and your ChatGPT sign-in.

The app stores the non-secret parts of a Credential locally, such as its name, username, authentication method, and private-key file path. A private key remains at the file path you selected. When a connection or provider request needs a secret, the desktop host resolves it from Windows Credential Manager; the UI does not display the saved value back to you.

Windows Credential Manager protects secrets at the Windows-account boundary. It does not protect a secret from malware or an attacker who has already compromised that account or computer. Secure the workstation, use device encryption and endpoint protection, and remove unused Credentials.

3. Understand SSH identity checks#

On first connection, the app asks you to confirm the SSH host-key fingerprint. Compare it with a value obtained out of band. After acceptance, the app stores the host identity for that saved device, so overlapping private addresses in different networks do not share one trust record.

If that host key later changes, AI NetShell refuses the connection and does not provide an in-app override. Investigate unexpected key changes before reconnecting.

Legacy SSH compatibility is an explicit per-device option for older equipment. Enable it only when the authorized device requires it, and prefer modernizing the device configuration.

4. Understand what an AI request contains#

An AI request is initiated by your action. Depending on the task, it can contain:

• Your question and instructions
• Conversation and tool context needed to continue the task
• Device output produced by the reads you triggered

AI NetShell applies recognized-secret redaction to device output before it is included. Redaction reduces exposure but cannot guarantee that every sensitive value, address, username, or business detail will be recognized. Review the scope of the requested read and avoid collecting unnecessary secrets.

AI NetShell does not perform a background upload of your device configuration as part of the Included AI workflow.

5. Choose the AI data route#

For the current legal description of data practices, use the Privacy Policy. Do not infer one provider's handling from another provider or from the Included AI statement.

6. Understand command and change safeguards#

Eligible read-only AI commands can run automatically when automatic reads are enabled. Reads that are not eligible for automatic execution stop for approval or are refused. Every AI-originated configuration change requires a separate human approval.

For a governed change, AI NetShell:

1. Checks the complete ordered proposal against the selected platform's write policy.
2. Captures a redacted before state and builds an exact current-versus-after preview.
3. Shows risk, recovery, and management-path information.
4. Requires the operator to approve the exact change, with an extra device-name confirmation for high-impact changes.
5. Applies to the bound target and records the outcome.
6. Runs available post-change checks and retains rollback only when the platform has a recovery method.

Rollback is not guaranteed. A lost management path, offline device, missing recovery snapshot, concurrent change, rejection, partial completion, or timeout can prevent remote recovery. See Change safety before enabling configuration changes.

AI NetShell lets only one of its own tasks type on a device session at a time. A change or a rollback holds the session from its first line to its last cleanup line, and the app's own checks and Lina's reads wait for it. Your own typing in that device's terminal is not held back, so wait until a change has finished before you type there.

7. Protect local records and exports#

The desktop keeps inventory, authentication-profile metadata, host identities, settings, audit records, investigation timelines, discovery approvals, snapshots, case notes, and exports in local application data or locations you select.

These records can contain device addresses, usernames, commands, configuration details, and operational findings even when saved secret values are kept separately. Protect the Windows profile and backups, limit access to exported reports, and redact diagnostic excerpts before sharing them. See Troubleshooting for the file locations and redaction checklist.

8. Verify updates#

AI NetShell downloads and verifies an app update before it stages the update for restart. It never restarts itself. Use Restart now only after the app reports Update ready.

The installer and every program file it installs, including AI NetShell.exe, are signed by PACKET 365 LLC, so Windows SmartScreen and your security tools can confirm the publisher of the app itself. See Install & update for the checks.

Expected Result#

You can identify which responsibilities stay in the desktop host, where saved secrets are resolved, what each AI route sends, which actions require approval, and where separate legal and security details live.

Troubleshooting#

A host-key fingerprint changed#

Do not reconnect until you have confirmed the device identity through a trusted channel. A changed key can be legitimate after replacement or rekeying, but it can also indicate redirection or interception.

A Credential may be exposed#

Rotate it at the device or provider first, then update or remove the saved Credential in AI NetShell. Review the relevant device and provider activity outside the app. Do not rely on deleting the local label as a substitute for rotation.

You are unsure which AI route is active#

Open the model picker and read the group heading. Included AI models are grouped under Included with your plan: uses credits. Anthropic-key and ChatGPT-account groups identify Your own AI routes.

Your organization needs a formal legal answer#

Use the Privacy Policy and Terms of Service as the routed legal sources. This product overview does not replace those pages or your organization's review.

Related#

• Credentials
• Set up AI
• Change safety
• Troubleshooting
• Known limitations