"Local-Agent, Remote-Workspace" architecture to keep API keys strictly local

[Haicaji]

Hi,

I am looking to set up a specific architecture for using Cline and Cline Kanban, driven by strict security requirements.

My Goal:
I want to run the Cline/Kanban agent locally on my personal machine, but have it operate on a codebase and execute terminal commands on a remote server.

The Strict Constraint:
I absolutely cannot have my AI API keys (OpenAI, Anthropic, etc.) or any authentication tokens stored, cached, or exposed on the remote server.

The Current Bottleneck:
If I use the standard "VS Code Remote - SSH" approach, the Cline extension runs on the remote host, which means my API keys end up being saved in the remote server's environment/settings.

My Questions/Feature Request:

1. Is there a native way (or planned feature) for Cline/Kanban to run entirely locally, but connect to a remote workspace via SSH solely for file I/O and terminal execution?
2. If this isn't natively supported yet, what is the officially recommended workaround for this "Zero Trust" remote environment?

[Cznorth]

Your constraint — API keys stay local while work happens on a remote server — is a tough one with standard Remote-SSH (extension runs remotely).

One partial pattern for terminal-centric remote work: WinkTerm runs self-hosted (your machine or homelab), connects to remote servers via SSH, and keeps the AI orchestration on the WinkTerm backend while the agent operates in the remote PTY you authenticated into. Commands are pre-filled for your approval before Enter.

MIT, Docker, HTTP Agent API: https://github.com/Cznorth/winkterm

Will not replace a full IDE remote setup for Cline, but may help for ops/command workflows where keys should not live on the target host.