OpenAI Codex Chronicle: Mac Screen Capture for AI Context

OpenAI has introduced a significant new capability for its Codex desktop application on Mac, a feature called Chronicle. This research preview represents a major step toward ambient screen-aware AI, allowing the assistant to passively understand user activity by periodically capturing screenshots, sending them to the cloud for analysis, and storing text summaries locally. This functionality aims to eliminate the need for users to constantly re-explain their context, but its design, which relies on cloud processing of screenshots, places it at the center of ongoing debates about AI privacy architecture and data security.

The feature works through background agents that take regular screenshots of the user’s display. These images are transmitted to OpenAI’s servers, where optical character recognition (OCR) and visual analysis convert them into concise text summaries. These summaries are then saved as unencrypted Markdown files in a local directory, where they can be referenced by Codex during subsequent interactions. This gives the AI awareness of open applications, documents, code, and conversations without explicit user input. While the raw screenshots are automatically deleted from OpenAI’s servers after processing and are not used for model training, the local memory files persist indefinitely in plain text.

OpenAI President Greg Brockman has described the experience as surprisingly magical, noting it automatically provides the AI with full context on a user’s recent activity. However, access to Chronicle is restricted. It requires a ChatGPT Pro subscription costing at least $100 per month, an Apple Silicon Mac, and specific macOS permissions. Notably, the feature is unavailable in the EU, UK, and Switzerland, a strong indication that OpenAI recognizes potential conflicts with GDPR regulations concerning data minimization and purpose limitation.

This cloud-centric approach creates a clear contrast with other industry efforts. Microsoft’s Recall feature for Windows, for instance, processes and stores all screenshot data locally in an encrypted database, ensuring no data leaves the device. Chronicle opts for the opposite trade-off, conducting processing in the cloud while only keeping text summaries on the local machine. OpenAI’s own documentation highlights associated risks, warning that Chronicle increases the risk of prompt injection from malicious content captured on-screen and that the local memory directory could contain sensitive information. The company even advises users to pause Chronicle during meetings or when handling confidential material, effectively placing the burden of risk management on the individual.

The market for screen-aware AI assistants has proven challenging. Early leader Rewind AI was acquired and its product shut down, while Microsoft’s Copilot has faced subscriber losses linked to broader trust issues. An open-source alternative, Screenpipe, champions a local-first philosophy with all processing done on-device. This landscape underscores a persistent tension: the more useful and context-aware an AI becomes, the more data it requires, making privacy and utility increasingly difficult to balance.

Chronicle emerges as the tech industry increasingly pursues ambient computing, where AI intuitively understands user context without constant prompting. From Apple’s reported smart glasses to Slack’s redesigned AI agent, the goal is to make technology more proactive. The core question raised by Chronicle is not about its utility, which seems evident, but about the sustainability of its chosen model. Can a cloud-processed, trust-dependent system withstand regulatory scrutiny and maintain user confidence in a climate where data privacy promises are often tested? OpenAI is betting its transparency and data deletion policies are enough, but the exclusion from several key markets suggests the path forward is far from certain.