Summary: OpenAI’s Codex for Mac added a research review feature called Chronicle, which periodically captures screenshots, sends them to OpenAI’s servers for processing, and stores text summaries as local unencrypted Markdown files to give the AI assistant passive context about user activity. This feature is not available in the EU, UK and Switzerland, requires a $100+ per month Pro subscription and Apple Silicon, and represents OpenAI’s first implementation of environment-aware AI on the desktop, opting for cloud processing and a helper instead of the native-first privacy architecture adopted by competitors such as ScreenwindeAI and now Rewin-AI.
OpenAI’s Codex desktop app for Mac has gotten a feature called Chronicle that periodically captures your screen, processes the content into text summaries, and stores those summaries as local memory files, giving the AI assistant context about what you’re working on. Released as research preview, the feature means that Codex can now understand your recent activity without explaining it. It also means that OpenAI sends screenshots of your desktop to its servers for processing, a design choice that puts Chronicle in direct tension with the privacy-first direction most of the industry is going.
Chronicle is part of a broader update that transforms Codex from a coding assistant to a general-purpose AI workspace. The April 16 release added “code for (almost) everything” PC usability, which allows Codex to run Mac programs with its own cursor, in-app browser, image generation, persistent storage, and more than 90 plug-ins. More than a million developers have used Codex, and usage has doubled since the release of GPT-5.2-Codex in December.
How Chronicle works
Chronicle runs background agents that periodically take screenshots of your screen. Those screenshots are sent to OpenAI’s servers, where they are processed using OCR and visual analysis to generate a text summary. Summaries are stored here as Markdown files in a local directory ~/.codex/memories_extensions/chronicle/. When you later access Codex, those memory files are inserted into its context window, allowing it to understand what programs you’re using, what documents you’re reading, what code you’re writing, and what conversations you’re having.
Raw screenshots are temporarily stored in the system temporary directory and automatically deleted after six hours. OpenAI states that screenshots are not stored on its servers after processing and are not used for training. And the memories created are stored indefinitely on your machine as unencrypted plain text files.
Greg Brockman, president of OpenAI, described the feature as “an experimental feature that allows Codex to see and have the most recent memory of what you’ve seen, automatically giving you the full context of what you’re doing. It feels surprisingly magical to use.”
Privacy architecture
Chronicle macOS requires Screen Recording and Accessibility permissions. It’s only available on Apple Silicon Mac devices running macOS 14 or later and only to ChatGPT Pro subscribers paying $100 or more per month. It is not available in the EU, UK or Switzerland, a geographic restriction that strongly suggests OpenAI recognizes the feature’s incompatibility with GDPR requirements regarding data minimization and purpose limitation.
The comparison with Microsoft Recall is instructive. Launched on Windows Copilot+ PCs, Recall takes screenshots every few seconds and stores them in an encrypted local database, with all processing done by the neural processing unit on the device. No screenshot data is output from the machine. Chronicle takes the opposite approach: processing happens in the cloud, but only text summaries are stored locally. Recall encrypts its database and requires biometric authentication via Windows Hello. Chronicle stores its memories as unencrypted Markdown files accessible to any process running on the computer.
OpenAI’s own documentation clearly acknowledges the risks. Chronicle “rapidly increases the risk of injection” because malicious content on a website you visit can be captured in a screenshot and interpreted as instructions by AI. The Memories folder “may contain sensitive information.” And the feature “uses rate limits quickly,” meaning Pro subscribers can find Codex usage limited by Chronicle’s background activity.
OpenAI recommends pausing Chronicle before meetings or when viewing sensitive content. Users can pause and resume through the Codex menu bar icon. The recommendation is self-revealing: it accepts that the feature will catch things it shouldn’t, and puts the burden on the user to manage that risk.
Category and its losses
Screen-aware AI assistants have had a turbulent history. Rewind AI, the most prominent early entrant before being acquired by Meta in December 2025, was rebranded as Limitless. The Mac application closed and disabled the screenshot. Microsoft copilot It lost 39% of its subscribers in 6 months, partly due to trust issues leading up to the Recall. A security researcher demonstrated in early 2026 that Recall’s encrypted database could still be used, fueling ongoing concerns since the feature was announced.
Open-source alternative Screenpipe offers a native-first approach: continuous screen and audio capture is handled entirely on-device, with a $400 lifetime license and no recurring cloud dependency. Perplexity PC the software takes another approach, turning the Mac mini into a persistent AI agent with access to local files and apps, though it also relies on cloud processing for its core intelligence.
The pattern across categories is consistent: the more useful a screen-aware AI becomes, the more data it needs to process, and the harder it becomes to reconcile that data appetite with user privacy. Chronicle favors a privacy architecture, betting that OpenAI’s promise not to store or teach data, combined with its six-hour deletion window, is enough to win user trust. Whether or not this gamble pays off depends entirely on whether users believe in the promise and whether OpenAI can maintain it at feature scale.
The surrounding computing context
Chronicle comes as the industry converges on the idea of AI assistants understanding your context without being told. Apple is experimenting AI smart glasses The environment for Apple Intelligence is designed as access channels. Slack’s recent AI overhaul It turned Slackbot into a deep contextual desktop agent about your business communications. OpenAI itself, with Jony Ive, is developing a display-less hardware device that is clearly positioned for the “ambient artificial intelligence” era. Gartner predicts that by 2026, more than 40% of large enterprises will deploy environmental intelligence pilots.
The thesis is that AI is dramatically more useful when it has passive, continuous access to what you do, rather than requiring you to express your needs from scratch every time. Chronicle is OpenAI’s first desktop application of that thesis, and it works: By Brockman’s account and feature design, eliminating the need to re-explain context to an AI assistant is a real productivity gain.
But the thesis has a price. Privacy-first alternatives Proton’s AI tools demonstrate that useful AI can run natively on open source models without sending user data to anyone’s servers. The question Chronicle asks is not whether screen-aware AI is useful. Obviously it is. The question is whether OpenAI’s chosen cloud-based, trust-dependent model will hold up in a regulatory environment that has already ousted it from three jurisdictions and users who watch enough AI companies promise data privacy to quietly revise their terms only when the economics demand it.
