OpenAI adds screen-aware Chronicle memory to Codex

Every time you turn to an AI assistant for help, the most tedious part is usually not the problem itself, but the context dumping. You have to explain what error popped up on your screen, which file you were just looking at, or what you were tweaking two weeks ago.

OpenAI is trying to kill that friction. The company just released Chronicle, a new memory feature for the Codex Mac app designed to automatically pick up context directly from your recent screen activity.

An agent watching over your shoulder

Chronicle is rolling out as a research preview for Codex Pro subscribers on macOS. It builds on the memory preview OpenAI introduced just last week when it launched the overhauled Codex desktop client.

Here is how it actually works. Chronicle runs background agents that take temporary screen captures to generate structured memories. These screenshots and the resulting memories are stored locally on your Mac, and you can inspect or edit what Codex remembers about you.

OpenAI says the goal is to make vague requests actually work. When you ask Codex about "this error" or "that doc," the app can lean on your recent visual history to figure out what you mean. Over time, it is designed to learn the tools you rely on and the projects you keep returning to.

To turn it on, users have to enable Memories in Codex settings, switch on Chronicle, and grant macOS Accessibility and Screen Recording permissions. You can pause or disable the feature at any time from the menu bar app.

That convenience comes with real trade-offs. OpenAI explicitly warned that Chronicle burns through rate limits very quickly under its current architecture. The company also noted that while files remain on device, other local applications on your machine may be able to access the stored memory files.

Why it matters

Codex is rapidly morphing from an agentic coding environment into OpenAI's planned desktop superapp for all kinds of builders.

Chronicle takes the assistant model beyond conversational chat and pushes it into ambient computing. Instead of forcing you to explicitly prompt the model with backstories, Codex passively observes how you work. For power users jumping between codebases and complex documentation, skipping the context rehash is a genuine quality-of-life win.

That said, we think most casual users should approach it with caution right now. The high token burn means you will likely hit your usage limits fast, and giving an app continuous screen recording access requires significant trust. Until OpenAI can optimize background token efficiency and ensure tighter local storage sandboxing, Chronicle remains an impressive preview of the future rather than an everyday necessity.