OpenAI has introduced "Chronicle," an experimental feature for its Codex application designed to provide the AI assistant with real-time context by observing a user's screen activity. This new capability aims to reduce the need for users to repeatedly explain their work, fostering a more seamless human-AI interaction. Greg Brockman, OpenAI's President, described the feature on social media, stating, > "Chronicle is an experimental feature giving Codex the ability to see and have recent memory over what you see, automatically giving it full context on what you're doing. Feels surprisingly magical to use."
Chronicle operates by running background agents that periodically capture screen content, process it into text summaries, and store these "memories" locally on the user's device. This allows Codex to understand elements like on-screen errors, open documents, or ongoing projects without explicit prompting. OpenAI emphasizes that this system helps Codex learn user workflows, frequently used tools, and recurring projects over time.
The feature is currently available as an opt-in research preview for Pro subscribers using the Codex app on macOS. To activate Chronicle, users must enable it within Codex's personalization settings and grant necessary macOS Screen Recording and Accessibility permissions. However, at its launch, Chronicle is not accessible in the European Union, United Kingdom, or Switzerland.
While enhancing user experience by providing passive context, OpenAI has outlined several considerations regarding Chronicle's use. The company notes that the feature can consume rate limits quickly and increases exposure to prompt injection attacks from on-screen content. Screen captures are temporarily stored, and generated memories are saved locally as unencrypted Markdown files, though users can inspect, edit, pause, or disable the feature at any time.
OpenAI CEO Sam Altman has also commented on the feature, likening its intuitive nature to "telepathy." This advancement signifies OpenAI's continued push towards more context-aware AI tools, aiming to transform human-computer interaction into a more collaborative and less instruction-dependent process.
