Codex Chronicle Goes Live - OpenAI's Assistant Now Reads Your Screen to Build Memory

OpenAI has released Chronicle, a research preview feature that enables its Codex assistant to build contextual memory by monitoring users' screens. Available today for ChatGPT Pro subscribers on macOS, Chronicle represents a significant expansion of OpenAI's memory capabilities beyond traditional chat history. The system operates by running background agents that periodically capture screenshots of users' displays. These images are sent to OpenAI's servers for processing through optical character recognition and visual analysis, then returned as plain-text memory summaries that are stored locally on the user's machine. The goal is to provide continuity between sessions, allowing Codex to understand ongoing projects, tools, and terminology without requiring users to restate context. However, the feature raises notable privacy and security concerns. Memory data persists unencrypted on local storage, whilst raw screenshot captures remain in temporary directories for up to six hours before deletion. The system also introduces new prompt injection risks, as screen content becomes a potential attack vector for malicious actors seeking to manipulate the AI's responses. Chronicle is currently unavailable in the EU, UK, and Switzerland, likely due to regulatory compliance considerations. Early users report that the feature consumes rate limits rapidly, suggesting significant computational overhead. Despite these limitations, the launch signals OpenAI's clear intent to develop ambient intelligence capabilities that extend far beyond conventional chatbot interactions.