Screenpipe — 24/7 Local Screen and Mic Recording for AI Agents

Introduction

Screenpipe continuously records your screen and microphone on your local machine, then indexes everything with OCR and speech-to-text. The resulting data is exposed via a local API that AI agents can query to understand what you have been doing, enabling context-aware automation without cloud dependency.

What Screenpipe Does

• Captures screen frames at configurable intervals with OCR extraction
• Records audio and transcribes speech to text locally
• Stores all data in a local SQLite database
• Exposes a REST API for searching and retrieving captured context
• Enables AI agents to act based on what you see and hear

Architecture Overview

Screenpipe runs as a background daemon written in Rust. A capture pipeline grabs screen frames and audio buffers, feeds them through OCR (via platform-native engines) and Whisper-based STT models, then stores structured results in a local SQLite database. A built-in HTTP server exposes search and retrieval endpoints. Plugins can subscribe to real-time events for immediate agent triggers.

Self-Hosting & Configuration

• Install via the one-line script or download binaries from Releases
• All data remains on your machine in ~/.screenpipe
• Configure capture resolution, FPS, and audio device in config.toml
• Set retention policies to auto-delete data older than N days
• Runs on macOS, Linux, and Windows

Key Features

• Fully local processing with no data leaving your machine
• Combined screen OCR and audio transcription in one tool
• Plugin system for triggering custom actions on captured events
• REST API for integration with any AI agent or automation
• Low CPU overhead through smart frame-diffing and batched processing

Comparison with Similar Tools

• Rewind.ai — macOS-only closed source; Screenpipe is open source and cross-platform
• Windows Recall — OS-level feature limited to Windows; Screenpipe works everywhere
• anarlog — meeting-focused note-taking; Screenpipe captures all screen activity continuously
• OpenRecall — similar concept but less actively maintained
• ActivityWatch — tracks app usage time; Screenpipe captures actual content via OCR

FAQ

Q: How much disk space does continuous recording use? A: With default settings (1 FPS, compressed), roughly 2-5 GB per day depending on screen resolution and audio duration.

Q: Does it impact system performance? A: Minimal. Frame diffing skips duplicate captures, and transcription runs in batched background threads.

Q: Can I query it from Claude Code or other AI agents? A: Yes. The REST API is agent-friendly and returns structured JSON with timestamps and content.

Q: Is the audio recording always on? A: You control which audio devices are recorded and can pause capture at any time.

Sources

• https://github.com/screenpipe/screenpipe
• https://screenpi.pe