Most "AI memory" tools ask you to feed them documents and notes; screenpipe takes the opposite bet — the richest context about your work already flows across your screen and through your microphone, so it just records all of it, locally. The wager is that a complete, queryable log of what you saw, said, and heard beats any hand-curated note system, and that the usual privacy objection disappears when nothing ever leaves your machine.
What Sets It Apart
- Event-driven capture, not a tape recorder. It grabs a screenshot only when something meaningful changes, then pairs it with the OS accessibility tree for clean text (OCR is the fallback). That keeps it around 5-10% CPU, which is the difference between a demo and something you actually leave on 24/7.
- Local-first, auditable by design. Everything lands in a full-text-searchable SQLite database with optional encryption at rest and no cloud round-trip. Your history stays yours, and the source-available code means you can verify that claim rather than trust it.
- A context source, not just a viewer. The "Pipes" plugin system runs scheduled AI agents over what you captured, and a built-in MCP server lets tools like Claude or Cursor query your timeline directly — turning your activity log into fuel for other AI.
Who It's For
Great fit if you want an open, self-hosted answer to Rewind.ai or Microsoft Recall, or if you're a developer who wants a local REST/MCP feed of your own screen and audio to build agents on. Look elsewhere if you need a polished, hands-off consumer app, or if always-on capture is a non-starter — recording everything raises real consent questions the moment someone else's data appears on your screen, and it trades storage and constant background work for that total recall.
