screenpipe

It creates a persistent, AI-searchable record of the user's entire digital activitydocuments read, conversations seen, video content watched, websites visitedenabling retrieval of anything the user has encountered, even if the original source is no longer accessible.

The system runs locally with all processing on-device: screen frames are OCR'd using local Tesseract or vision model inference, audio is transcribed using Whisper running locally, and the resulting text is indexed in a local vector database for semantic search.

An API layer allows AI assistants and plugins to query the accumulated knowledge baseenabling use cases like 'find that article I read last week about X' or 'what did that meeting participant say about Y last month' through natural language queries against the personal activity history.

Knowledge workers managing information overload who want to retrieve things they've seen without relying on memory or bookmarking, researchers who need to cite sources they encountered but didn't save, and privacy-conscious users who want the utility of AI memory without sending their screen content to cloud services use Screenpipe as a personal AI memory layer.

Its local-first architecture addresses the significant privacy concern of screen recording tools that upload activity data to cloud servers, keeping all captured content and queries on the user's own hardware.