Is screenpipe free to use?

The source is available for personal, non-commercial use under the Screenpipe Commercial License. The signed desktop app requires a paid subscription (Standard tier starts at $25/month), and commercial use requires a separate license.

Does screenpipe send my data anywhere?

No, by default all screen and audio data is stored locally in a SQLite database on your device. Cloud transcription (Deepgram) and cloud AI models are optional and off by default.

What platforms does screenpipe support?

macOS (Apple Silicon and Intel) and Windows 10/11 have full native installer support. Linux is supported but requires building from source.

How does screenpipe give context to AI coding assistants?

It runs an MCP server that tools like Claude Desktop, Cursor, VS Code (Cline, Continue), and other MCP-compatible clients can query for recent screen and audio history, plus a REST API and SDK for custom integrations.

Can I control what data AI agents (pipes) can access?

Yes, each pipe is a markdown file with YAML frontmatter that restricts allowed/denied apps, windows, content types, time ranges, and API endpoints. This is enforced deterministically at the skill, agent, and server level, not through prompting.

How does screenpipe compare to Rewind or Microsoft Recall?

Unlike those, screenpipe is source-available and cross-platform (macOS, Windows, Linux via source), stores data 100% locally by default, records all monitors, and offers a developer API and plugin system for building custom agents.

screenpipe: Local-first screen and audio

Screenpipe is a source-available application, written in Rust, that continuously captures your screen and audio locally, builds a searchable history, and exposes it to AI agents. It's aimed at knowledge workers, developers using AI coding assistants, researchers, and teams who want an auditable alternative to cloud-based memory tools like Rewind or Microsoft Recall.

Instead of recording video frames constantly, it listens for events (app switches, clicks, typing pauses) and captures a screenshot plus the OS accessibility tree when something changes, falling back to OCR when accessibility data isn't available. Audio is transcribed locally with Whisper or via Deepgram, with speaker diarization. Everything lands in a local SQLite database that's searchable via natural language, full-text search, or raw SQL, and reachable by AI assistants through an MCP server or REST API.

Key features

Event-driven capture: screenshots and accessibility-tree text are captured only on meaningful changes, keeping CPU use around 5-10% and storage around ~20GB/month
Audio transcription: local Whisper (Large-V3-Turbo) or Deepgram transcription of system audio and microphone, with speaker identification, works with Zoom, Meet, Teams, and any audio source
AI-powered search: natural language search across OCR text, accessibility text, and transcripts, filterable by app, window, URL, and date range
Timeline view: a scrollable DVR-style view of your screen history with linked audio playback
Pipes (plugin system): scheduled AI agents defined as markdown files that query your screen data, call APIs, and take actions; built-in pipes cover meeting summaries, day recaps, standups, and time breakdowns
Per-pipe data permissions: YAML frontmatter controls which apps, windows, content types, time windows, and endpoints a pipe can access, enforced at the OS/skill/server level rather than by prompting
MCP server: exposes screen history to Claude Desktop, Cursor, VS Code extensions, and other MCP-compatible clients with a single install command
Developer API: full REST API on localhost, plus a JavaScript/TypeScript SDK and raw SQL access to the underlying database
Privacy controls: filters for windows, apps, browser extensions, passwords, and a proprietary PII model, plus optional encryption at rest, all running fully offline

Ideal use cases

Giving Claude Code, Cursor, Cline, or Continue live context about what you've been working on, so they can reference recent code, docs, or conversations without you re-explaining
Recalling what was said in a meeting, what a document said, or which tab you had open three hours ago
Automated meeting notes, standup updates, or daily recaps built from your actual screen and audio activity via pipes
Teams that want a shared "company brain" with deterministic, per-pipe data permissions instead of blanket surveillance
Researchers or people with ADHD who lose track of open tabs, documents, and half-finished threads

It's not a fit if you need a cloud-synced memory tool across devices without local storage, want a fully free commercial deployment (the signed desktop app requires a paid subscription for non-personal use), or you're on Linux and need a polished installer rather than a source build.

Installation

The simplest path is the desktop app, which includes auto-updates and all features:

download from https://screenpipe.com/how-to-install?download=1

Alternatively, run the CLI directly with npx:

npx screenpipe record

Then set up integrations and skills:

npx screenpipe setup
# or, for Claude specifically
claude mcp add screenpipe -- npx -y screenpipe-mcp@latest

Once set up, you can ask Claude things like "what did I see in the last 5 mins?" or "summarize today's conversations," or have it create a pipe that automates a task based on your activity.

For coding agents (Claude Code, Codex, Gemini CLI, Cursor) working from this repo directly, point them at the CLI skill file (crates/screenpipe-core/assets/skills/screenpipe-cli/SKILL.md) and run npx screenpipe setup to install skills and MCP configuration for every detected agent.

Platform support: full native installers for macOS (Apple Silicon and Intel) and Windows 10/11; Linux requires building from source. Minimum recommended specs are 8GB RAM, 5-10GB disk per month, and modest CPU overhead thanks to event-driven capture.