Windows Recall drew intense public scrutiny when it launched, mainly because its initial implementation stored screenshots of every screen in an unencrypted, easily-accessible database — a design flaw that was discovered almost immediately after the feature’s announcement and forced Microsoft to delay its rollout for substantial redesign.
In April 2026, OpenAI announced Chronicle, a feature that does something architecturally similar: it takes periodic screenshots of your screen to give the AI context about what you’re working on.
Chronicle received significantly less scrutiny than Recall. This is worth examining, because the underlying questions are similar, and the specific implementation details that make the two products different don’t make Chronicle’s privacy implications any less significant.
What Chronicle Is
Chronicle is an AI memory feature for OpenAI’s Codex product — OpenAI’s coding-focused AI assistant. The premise is straightforward: by periodically capturing your screen, Chronicle can understand your context. What files do you have open? What error message appeared five minutes ago? What were you working on before you opened this new file? This context helps the AI give more relevant, accurate assistance than it could from a single, explicit prompt.
The feature is framed as making Codex smarter by giving it access to your working context — similar to how a human collaborator would naturally absorb context by watching over your shoulder, rather than being briefed from scratch each time you ask a question.
From a product perspective, the appeal is obvious. AI assistants that know what you were just doing are genuinely more useful than ones that don’t. The friction of having to describe your context every time you want help is real, and features that eliminate it feel like improvements.
From a privacy perspective, the mechanism for achieving this — taking screenshots of your screen at intervals — is the same mechanism that made Recall controversial. What’s captured and where it goes are the questions that determine whether “more useful AI assistant” and “more comprehensive record of your activity” are acceptable tradeoffs.
How Chronicle Differs From Windows Recall
The two features are similar in concept and different in important specifics.
Recall is local; Chronicle is cloud-connected. Windows Recall processes screenshots on your device, stores them in a local database, and (in its current form) doesn’t upload screenshots to Microsoft’s servers. Chronicle, as part of OpenAI’s cloud product, sends data to OpenAI’s infrastructure. Screenshots or contextual data derived from them leave your device.
This is a fundamental architectural difference. Recall’s risks are primarily about what an attacker could do with access to your local machine — whether through malware, physical theft, or a court order compelling device access. Chronicle’s risks include all of those plus the risk that data on OpenAI’s servers is exposed in a breach, subpoenaed from OpenAI directly, or used in ways you didn’t anticipate when you opted in.
Recall is in Windows by default for qualifying hardware; Chronicle requires opt-in to Codex. To use Chronicle, you need to be using Codex and actively enable the feature. The user population is smaller and, at least initially, more technically aware.
Recall captures everything on screen continuously; Chronicle captures to provide context for coding work. Chronicle’s stated scope is narrower — it’s designed to capture coding-related context, not everything you do on your computer.
The “narrower scope” caveat is important, though: when you’re writing code, your screen may show other applications, personal data, test data in development databases, credentials in configuration files, or communication windows with colleagues. A feature that captures your screen to understand your coding context captures everything visible during that context.
What Chronicle Captures (And What That Includes)
OpenAI has been deliberately vague about the specific technical implementation of Chronicle — whether it takes literal screenshots, extracts text from the screen, or uses another mechanism. The practical question, from a privacy standpoint, is what information about your screen becomes available to OpenAI’s servers.
Based on the stated functionality — giving Codex context about what you’re working on — the system needs to capture, at minimum:
- The content of files you have open. Source code, configuration files, documentation, and anything else in your editor or terminal.
- Error messages and terminal output. Chronicle is most useful for capturing the context that’s hard to copy-paste into a prompt: error messages as they appeared, the output of a command that failed, what was on screen when a bug manifested.
- Adjacent windows and applications. If you switch between your editor and a browser showing documentation, or between your editor and a Slack window discussing the code, Chronicle’s screenshot mechanism captures whatever is visible.
That third category — adjacent content — is the one most likely to contain personal or sensitive information that has no direct relevance to the coding task. A credentials file briefly visible in a file browser. A personal message in the corner of the screen. Financial data in a spreadsheet open in the background. Test data that contains real user information (a common and problematic practice in development environments).
This isn’t specific to Chronicle — it’s a structural issue with any “capture what’s on screen” feature. Screens contain more than the task at hand.
The Trust Question: What OpenAI Does With Screen Data
When a feature sends screenshot data to a cloud service, the key questions are:
Is the data used to train AI models? OpenAI’s policies on using user input for model training have evolved and are context-dependent. Business subscribers with certain API arrangements have different defaults than consumer product users. Chronicle’s specific data retention and use policies, at the time of writing, haven’t been published in the level of detail that would resolve this question clearly. This is worth watching.
How long is data retained? A screenshot taken to provide context for a coding session five minutes ago is useful immediately and of no value afterward. What happens to it after that period? Is it retained for potential future reference, for debugging, for product improvement? Retention policy matters because data that’s retained is data that can be breached, subpoenaed, or repurposed.
Can the data be accessed by OpenAI staff? Cloud storage means someone at OpenAI could access it, subject to their internal policies and access controls. This is different from local storage where access requires physical or privileged access to your specific machine.
What happens if OpenAI is acquired, goes through bankruptcy, or changes its policies? Data collected under one policy regime can persist through company transitions that change how it’s treated. This is true of any cloud service.
These questions don’t have definitive public answers for Chronicle at the moment. That gap is itself informative: a feature that takes screenshots of your screen and sends them to a cloud service should have clearly published answers to these questions before asking users to enable it.
The Pattern: AI Memory Requires Continuous Capture
Chronicle and Recall share a design pattern that’s becoming common across both OS-level and application-level AI features: to make AI assistants more contextually aware, they need to capture more about what you’re doing, more continuously, and either process it locally or send it somewhere with more compute than your device.
This is sometimes called “AI memory” — the idea that an AI assistant should remember context between sessions, learn your preferences over time, and understand your current working environment without being explicitly briefed. It’s a genuine product improvement direction, and many users find the resulting experiences more useful.
But “AI memory” requires storing information about you somewhere — either locally, which creates risks around device access, or in the cloud, which creates risks around server-side access. There’s no version of “AI that remembers your context” that doesn’t involve a record of your context being created and stored.
The design choices that matter: Where is the data stored? For how long? Who can access it? Is it used to train models? Can it be deleted? What is it used for beyond the immediate feature?
Products that are explicit about these answers make it possible to evaluate the tradeoff. Products that are vague leave users making a decision with incomplete information.
Recall, Chronicle, and the Next Wave of Features
It would be convenient if Chronicle and Recall were the end of this design pattern, and the questions they raise were answered for the set of known features. They’re not.
AI features that improve through contextual awareness are being added across productivity software, IDEs, browsers, and mobile operating systems. The underlying mechanism — some form of monitoring what you’re doing to give AI context — is generic, and it will be applied to many different feature ideas by many different companies. Some will be more privacy-respecting than others. Many will make the same tradeoff: useful AI context in exchange for a persistent record of activity.
The questions to ask of each feature are the same:
- Does this send data to a cloud server?
- What data specifically?
- How long is it retained?
- Who can access it?
- Is it used to train models?
- Can I delete it?
- What’s the default — opt-in or opt-out?
These questions apply to Chronicle, to future iterations of Recall, to browser AI assistants that summarize your browsing history, and to any similar feature that derives its value from monitoring your activity over time.
What Deliberate Storage Looks Like Instead
The contrast is worth stating directly: a tool like daftei is the opposite design philosophy.
Nothing happens automatically. No screenshots, no background monitoring, no activity history generated by the tool itself. You decide what to upload — a photo, a document, a voice note — and it’s stored encrypted in transit with TLS 1.3 and at rest with AES-256. The app doesn’t learn from your uploaded files to personalize your experience. It doesn’t pass your content to third-party AI. It doesn’t build a record of what you were doing when you weren’t using it.
That’s not because automation is inherently bad. It’s because the right model for personal memory — your own photos, your own documents, your own records — is one where you choose what to save, rather than one where the system captures everything and lets you search it later.
AI context features and personal memory tools are solving different problems with different tradeoffs. Being clear about which problem you’re solving with which tool helps avoid the situation where you’ve enabled “AI context” features across your environment and later discover that the record of your activity you’d most like to be private is exactly the record those features were built to maintain.
Chronicle is worth watching. The feature as described is genuinely useful for the context it’s designed to serve. The questions about what OpenAI does with the screen data it collects through Chronicle deserve answers before you decide to use it — not after.