screenpipe
A local, continuous screen-and-audio recorder that gives AI coding agents such as Claude, Codex and Hermes searchable context about what you have actually done on your computer.
screenpipe records a computer's screen and audio continuously and locally, then uses that recording to give context to agents such as Claude, Codex, OpenClaw and Hermes. It captures the full accessibility tree, with OCR as a fallback, transcription, speaker identification, keyboard input and app switches, and stores everything only on the user's own machine.
A desktop app offers all features with auto-updates, or a CLI can run capture headlessly and register the project's skill and MCP configuration into detected coding agents, after which an agent can be asked things like what appeared on screen in the last few minutes, or told to build an automation triggered by particular activity. It runs at roughly 5 to 10 percent CPU and 0.5 to 3GB of memory, uses about 20GB of storage a month, supports filters for windows, apps, extensions and passwords, and offers optional encryption at rest.
It is source-available, works fully offline, and is written in Rust.