Use GPT through an SSH server

3 minute read · Chats & Local Agents

For accounts with GPT through SSH enabled, Vortex can run the supported Codex integration on your Linux server over SSH. This is a separate provider connection from a local coding-agent profile, ordinary SSH terminal and paired phone.

• Check availability and connect
• Choose the remote account in chat
• Handle interruption
• Verify the selected execution path

Check availability and connect#

1. Check that the Vortex account has the remote-agent capability enabled. If the SSH option is absent or access is denied, this account cannot use it yet.
2. Save and verify the server in SSH connections, including the server's host key.
3. In Settings → AI providers, add an OpenAI / GPT account with the SSH connection option. Give it a recognizable account name and choose the saved host.
4. Follow server preparation. Vortex reuses a ready worker or prepares the required remote components and checks the Linux sandbox. A failed connection or sandbox check must be resolved before proceeding.
5. Complete the displayed provider sign-in through its official page. Enter any code there, then return to confirm the connection.

Advanced settings allow an OpenSSH destination, Node executable and worker path. Prefer the prepared defaults unless the server has a deliberate custom setup. Installing missing system dependencies can require the server administrator; a saved SSH connection does not grant sudo rights.

Choose the remote account in chat#

Select the account marked GPT · SSH in the provider/account controls, then a model and reasoning level reported by that server's provider. Confirm the selected project and working context before asking for changes. Local and remote provider sessions are separate even when their account labels resemble each other.

The agent connection runs over SSH. Supported Vortex tools and computer-control actions remain subject to the desktop conversation's resource boundaries, system permissions and access mode. Using a Linux provider host does not give it unlimited access to the desktop or make an ordinary local project into a synchronized remote checkout.

Handle interruption#

Keep the SSH connection and Vortex available. Stop, disconnection and loss of account permission end active access. Inspect the conversation and changed files before retrying; a disconnected result can follow a partially completed operation.

Host-key checks remain part of SSH authentication. Keep SSH credentials in the encrypted connection settings and provider sign-in in its supported flow. Do not paste either into the chat. Remote account availability is separate from a provider's own model eligibility and usage limits.

Verify the selected execution path#

Start with a small read-only request, confirm the reported account and context, then make one bounded change and review its result. The current worker integration is for Codex; the SSH option does not imply equivalent remote execution for every provider.

Related guides: SSH hosts, provider models, Computer use, permissions.