Saving up to 10 frequently used prompts
OpenAI updated the command-line version of its programming tool "Codex" to 0.164.0 on Oct. 23 (Oct. 24 Japan time). Thibault Sottiaux, who leads Codex, outlined the main changes on X.
The new "presets" let users save up to 10 frequently used prompts and commands, numbered, and recall them from a keyboard menu. Saved content persists the next time the tool is launched. Users can also now specify the model, reasoning depth, speed and permissions on the same line as a prompt. Previously, they had to enter a settings command before giving the prompt.
Signing in to enterprise MCP servers
Sign-in to enterprise servers using MCP, a standard for connecting external tools to AI, is now officially available. Users can choose from three methods: using the permissions of their Codex account, signing in through the server's own authentication system, or an OAuth method that requires a secret key. In terminals that support images, images are now displayed directly in the conversation history. Guidance on settings for using Amazon's Amazon Bedrock is also now shown from the start.
Usage has doubled since DevDay
Sottiaux said use of the command-line version has doubled since DevDay on Sept. 29. He also said the tool was changed so that changing the reasoning depth setting in the middle of a conversation no longer invalidates the cache. The cache is a mechanism that reuses processing of the same preceding content, and when it is invalidated, usage limits are consumed faster.
