llm-keys-ui 0.1 is a new plugin that provides a web interface for saving API keys to a machine without pasting them directly into agent sessions. It is designed for use with Codex Remote and LLM projects, allowing keys to be retrieved later via commands like `llm keys get anthropic`.
SynthePulse Insight · AI deep reading
Version 1 · 1 source
Simon Willison released llm-keys-ui 0.1, a plugin for setting LLM API keys in a web interface. Its starting point is not feature expansion, but avoiding one specific action: don't paste keys into an agent session.
The release notes for llm-keys-ui 0.1 come right out and acknowledge that the plugin "solves a very specific problem." It is not packaged as a general-purpose key management solution, but as an LLM web UI whose purpose is to set keys in an interface.
The problem comes from the author's own workflow: he started using Codex Remote to run coding agents on multiple machines and control them from his phone. When those machines are used for LLM-related projects, API keys occasionally need to be configured—and that is exactly the step this plugin handles.
The reason the author gives is a personal preference, not the conclusion of a security audit: he doesn't like pasting API keys into agent sessions. So he wanted a path that could get keys onto the machine without pasting them directly through the ChatGPT app.
Here it is necessary to distinguish fact from inference. Not liking to paste keys, and wanting to bypass pasting directly through the ChatGPT app, are motivations the author states in the release notes and count as source claims; whether this practice objectively reduces the risk of key leakage is not verified or assessed by the source and cannot be treated as a confirmed conclusion.
The usage given in the release notes is brief. The author can have Codex run uvx --with llm-keys-ui llm keys-ui --all, then have Codex tell him the access URL; that URL contains a LAN or Tailscale device IP and points to an interface for saving additional API keys.
After the key is written, it can later be retrieved in shell commands using something like llm keys get anthropic. The whole flow splits "writing the key" and "reading the key in a command" into two actions, with the web interface handling the input step in between.
The information currently confirmable is limited to the release notes themselves: version number 0.1, release date September 20, 2026, the plugin's positioning, the usage command, and the URL form. The source does not provide how keys are stored, encryption mechanisms, access control details, or any third-party verification or security assessment.
Therefore, understanding it as "a tool for a usage habit that avoids pasting keys into agent sessions" is supported by the source; understanding it as a "verified security solution" goes beyond the source's scope. The latter can only be kept as an open question to be verified.
This article is based entirely on a single source: Simon Willison's release notes for llm-keys-ui 0.1, published on September 20, 2026. That source is the author's own introduction to a plugin he developed, making it a first-party statement, but the parts concerning motivation and user experience remain subjective self-reporting; the source provides no technical details on key storage, encryption, or access control, and there is no third-party verification. Any judgments of the "reduces risk" kind are not adopted here, because the source does not support them.
The value of llm-keys-ui 0.1 is not in new capabilities, but in moving "configuring API keys" out of the agent session and into a web interface reachable over LAN or Tailscale; whether it is more secure is not answered by the source, which only answers how the author doesn't want to do it.
Primary report
Primary source