Simon Willison Launches llm-keys-ui for Remote Agents

Independent developer Simon Willison has launched llm-keys-ui 0.1, a specialized plugin aimed at resolving a common security headache for developers managing remote AI agents. The tool addresses the risks associated with configuring API keys on remote machines, particularly when controlling those systems through mobile devices or chat-based interfaces.

The plugin is designed for workflows where developers use tools like Codex Remote to run coding agents on external hardware. Previously, setting up necessary API keys for these remote LLM projects often required pasting sensitive credentials directly into agent sessions or chat applications like ChatGPT. To bypass this insecure practice, llm-keys-ui allows the remote agent to spin up a local web interface.

Once activated, the plugin generates a URL—which can resolve to a local network address or a Tailscale device IP—providing a secure interface where the user can input and save their API keys. Once stored, the keys can be retrieved programmatically on the remote machine. For example, a developer can run a command like "llm keys get anthropic" within a larger shell command to authenticate services on the fly.

For AI practitioners, this release offers a practical method to maintain credential hygiene while leveraging autonomous coding agents. By keeping API keys out of chat logs and agent prompt histories, developers can safely run remote experiments without exposing their private credentials to third-party LLM providers or intermediate logging systems.