recently, openai launched a new feature called chronicle for its codex application, which uses screen recording to build a contextual memory repository. this allows codex to automatically reference the user’s previous work in subsequent tasks, eliminating the need to restate instructions each time.
according to the introduction, chronicle runs continuously in the background, with an ai agent summarizing the screen-recorded content and saving it locally as a markdown file. the recorded files are stored only temporarily and are automatically deleted after six hours. this feature is designed to help codex better understand what the user means, which tools they’re using, and what project they’re working on, thereby improving collaboration efficiency.
currently, chronicle is available as an optional preview version for chatgpt pro subscribers on the macos platform, but it is not yet accessible in the european union, the united kingdom, or switzerland. users can enable the “memory” and “chronicle” features in the “personalization” settings of codex and grant macos permissions for screen recording and accessibility services to activate the feature.
openai also cautions users that chronicle can quickly consume call credits and may increase the risk of prompt-injection attacks—malicious commands could be injected through displayed web pages. in addition, the memory content is stored on the device in an unencrypted format, so users should exercise caution when using it. as this feature is still in the preview stage, openai states that it will continue to optimize it based on user feedback.
