Louis, founder of Screenpipe, has launched a new application that records user screens and audio locally, creating a searchable memory for AI agents. This aims to automate repetitive tasks and generate Standard Operating Procedures.

Screenpipe evolved from Louis's long-standing interest in building a "second brain" for himself, experimenting with RAG and AI plugins like Ava and Embedbase. He found that AI agents need context about user activity to be truly autonomous, moving beyond fine-tuning and tool-calling to a more integrated approach.

Instead of continuous video recording and OCR, Screenpipe now listens for system events like app switches and clicks, pairing screenshots with accessibility tree data. Audio is captured and transcribed locally. Data is stored in a local SQLite database and other files, with an AI-friendly API available for agents. Users can leverage this memory for tasks like gathering context, retrieving information, creating wikis, and automating workflows. Privacy features include PII redaction and filtering options. The codebase is primarily in Rust, with commercial use requiring a license, while personal and non-commercial use remains free.