OpenAI has added a feature called Computer History to its macOS desktop app that records your clicks, keystrokes, and unfinished tasks, building a persistent activity timeline that ChatGPT and Codex can reference when you make requests.
Key takeaways
Computer History is a macOS-only feature in the ChatGPT desktop app that logs user clicks, keystrokes, and incomplete tasks into a timeline.
Both ChatGPT and Codex can query the timeline, letting the AI pick up half-finished work or suggest automations based on observed patterns.
The feature turns your activity into training-adjacent context — a meaningful shift for creators who use the desktop app for creative workflows.
No equivalent opt-in is yet available on Windows or in the web interface, limiting the rollout for now.
Creators using ChatGPT for prompting or file work should review what the feature captures before relying on it in professional or client contexts.
ChatGPT's macOS desktop app now ships with Computer History, a feature that logs user activity into a persistent timeline.
Computer History works by monitoring what you do on your Mac — which apps you open, what you type, where you click — and stitching those actions into a chronological record. According to The Verge, that timeline is then available to ChatGPT and OpenAI's Codex agent, so either can resume a task you left half-done or suggest automations based on patterns it has observed.
What the timeline actually captures — and why the scope matters
For AI-art creators who use ChatGPT's desktop app to manage prompts, batch-rename outputs, or run Codex scripts against local image folders, the practical reach of this logging is wide. If you were mid-way through building a prompt library and closed the app, Computer History gives ChatGPT enough context to ask whether you want to continue — or to pre-fill a request with your last known state. That is genuinely useful friction-reduction.
The other side of that coin is that the feature observes everything in scope, not just ChatGPT-specific activity. Clicks and keystrokes across apps are fair game, which means any creative tool running alongside ChatGPT — Photoshop, a local diffusion UI, a file manager full of client assets — could contribute to the timeline. Creators working under NDAs or with client-owned assets should treat this as a data-handling question, not just a convenience feature, before enabling it.
This is not the first time an AI platform's ambient-capture ambitions have outrun user awareness. When Recall — Microsoft's analogous screenshot-based memory system — launched in 2024, the backlash over its always-on recording forced a rapid redesign and an opt-in gate. OpenAI has not yet detailed whether Computer History is on by default or what the retention window looks like, details that will matter enormously to creators who run sensitive projects through the desktop app.
Codex as the downstream beneficiary
The tighter integration with Codex is where this feature has the most direct workflow impact. Codex, OpenAI's coding-focused agent, can already execute multi-step tasks — running scripts, editing files, calling APIs. Giving it a history of your recent actions means it can make better-informed guesses about what you actually want when you issue a vague instruction like "finish what I was doing with the image batch."
For creators who use Codex to automate repetitive image-processing tasks — resizing outputs, tagging files, pushing assets to a folder structure — a context-aware agent that knows your last session state is a real productivity gain. The question is whether the tradeoff in ambient data collection is one each creator is comfortable making explicitly, not by default.
OpenAI has not announced a Windows rollout or a web-interface equivalent. The feature is currently macOS-only, which means the majority of ChatGPT's user base cannot access it yet — and gives OpenAI time to refine the privacy controls before a broader release. Those controls, and specifically whether Computer History data feeds model training, are the detail worth watching in OpenAI's forthcoming documentation.
