Perfect Recall is Microsoft’s experimental context window feature that provides virtually unlimited conversation memory by continuously storing and indexing your PC activity. Unlike traditional AI assistants limited to 128K or 200K tokens, Perfect Recall uses local storage to create a searchable timeline of everything you do, giving AI models access to months of contextual information.

How Large Is Perfect Recall’s Context Window?

Perfect Recall doesn’t use a traditional token-based context window. Instead, it captures screenshots every few seconds and stores them locally with OCR-indexed text. This approach provides effectively unlimited context—limited only by your storage capacity. A typical setup can retain 3-6 months of activity data, representing millions of tokens worth of information that traditional context windows couldn’t handle.

What Are Perfect Recall’s Key Limitations?

Privacy remains the primary concern. All data stays local on Copilot+ PCs with NPU chips, but the comprehensive surveillance aspect makes users uneasy. Microsoft has added encryption and sensitive content filtering, but the feature requires Windows 11 and specific hardware. Additionally, retrieval speed decreases as your timeline grows beyond 2-3 months of continuous capture.

How Does It Compare to Standard AI Context Windows?

GPT-4’s 128K token window holds roughly 300 pages of text. Claude 3 offers 200K tokens. Perfect Recall sidesteps these limits entirely through persistent storage rather than active memory, making it fundamentally different from—and potentially more powerful than—extended context windows in cloud-based models.