Screenpipe – Record Your Screen and Feed It to AI Agents

TL;DR

TL;DR: Screenpipe is a local-first screen recorder that captures everything you see and say, stores it on-device, and makes it queryable by AI agents via natural language. Install with npx screenpipe record and connect it to Claude, Cursor, or any MCP-compatible agent.

Source and Accuracy Notes

⚠️ This section is MANDATORY. All links must be verified from actual source, not guessed.

• Project page: screenpipe.com — verified via direct visit
• Source repository: github.com/screenpipe/screenpipe — README read in full
• License: Screenpipe Commercial License (source-available; personal/non-commercial use free, commercial use requires a license) — verified via LICENSE.md on main branch
• HN launch thread: news.ycombinator.com/item?id=49024620
• YC batch: S26 (verified via README and GitHub topics)
• Stars: 20,504 (verified via GitHub REST API)
• Source last checked: 2026-07-25

What Is Screenpipe?

Screenpipe records your screen continuously — every app you open, every page you scroll past, every call you take — and keeps that recording private on your own machine. It then transcribes the audio (your microphone and system audio) using Whisper running locally, and indexes everything so you can search it with natural language queries.

The core pitch from the README:

screenpipe turns your computer into a personal AI that knows everything you’ve done. record. search. automate. all local, all private, all yours.

The target user is someone who wants AI agents to have real context about what they’re working on — not just the current file, but the Slack conversation from yesterday, the Figma design from last week, or the meeting notes from this morning.

Setup Workflow

Step 1: Install

The fastest way to get started is the CLI via npx:

npx screenpipe record

This launches screenpipe in record mode. It will ask for screen recording permissions on macOS or Windows.

For a full desktop app with auto-updates, download from the onboarding page:

open https://screenpi.pe/onboarding

Step 2: Configure Initial Settings

Screenpipe stores data locally in ~/.screenpipe/ by default. On first run it creates the directory structure and starts capturing.

On macOS 14.4+, you can exclude specific apps from system audio capture by creating ~/.screenpipe/audio-exclusions.json:

{
  "excluded_apps": [
    { "bundle_id": "com.spotify.client", "name": "Spotify" }
  ]
}

Step 3: Connect an AI Agent

Screenpipe ships with MCP server support. You can wire it into Claude, Cursor, or any MCP-compatible agent:

npx screenpipe setup
# or with Claude MCP explicitly:
claude mcp add screenpipe -- npx -y screenpipe-mcp@latest

Once connected, you can ask natural language questions:

what did i see in the last 5 minutes?
summarize today's conversations
create a pipe that updates linear every time i work on task X

Step 4: Browse the Web UI

Screenpipe includes a local web interface for searching your history without an AI agent:

npx screenpipe browse
# or open directly
open http://localhost:3030

Deeper Analysis

How the Capture Works

Screenpipe uses event-driven capture instead of recording every frame. It listens for meaningful events — app switches, mouse clicks, typing pauses, scrolling — and only captures a screenshot when something actually changes. Each screenshot is paired with the OS accessibility tree (structured text describing buttons, labels, text fields), falling back to OCR when accessibility data is unavailable (e.g. games, remote desktops).

This approach keeps CPU usage at 5–10% on modern hardware and storage at roughly 20 GB per month of active use.

Audio Transcription

Screenpipe captures both system audio (what you hear through speakers) and microphone input (what you say). Transcription runs locally using Whisper Large-V3-Turbo, with an optional Deepgram cloud fallback. Speaker identification and diarization are included.

PII Filtering

Released May 2026, Screenpipe includes an on-device PII detection model that outperformed Google, Microsoft, and OpenAI models on computer recording data, running at 9ms on consumer hardware. This detects passwords, credit card numbers, and other sensitive data in real time as captures are written to disk. Optional encryption at rest is also available.

Platform Support

| Platform | Status | |----------|--------| | macOS (Apple Silicon) | Full support | | macOS (Intel) | Full support | | Windows 10/11 | Full support | | Linux | Build from source |

Minimum requirements: 8 GB RAM, ~5–10 GB disk space per month.

How It Compares

Screenpipe is the leading source-available alternative to Rewind.ai (now Limitless), Microsoft Recall, Granola, and Otter.ai. The key differentiator is that everything stays on your machine — no cloud upload, no subscription required for personal use.

Practical Evaluation Checklist

• [ ] Installed via npx screenpipe record
• [ ] Granted screen recording permissions
• [ ] Verified recordings appear in ~/.screenpipe/
• [ ] Asked a natural language query via MCP-connected agent
• [ ] Tested audio capture with a video call
• [ ] Checked exclusion filters for sensitive apps

Security Notes

• All data is stored locally by default — nothing is uploaded unless you explicitly configure cloud sync
• PII model runs on-device at inference time
• Optional encryption at rest available
• Source code is auditable (Screenpipe Commercial License)

The license does restrict commercial use — organizations using Screenpipe in a production/business environment need a commercial license. Personal, educational, and non-profit use is free.

FAQ

Q: Does Screenpipe record everything all the time? A: Yes, it runs continuously in the background. You can exclude specific apps from audio capture via audio-exclusions.json. PII filtering runs on screenshots as they are written, redacting sensitive content.

Q: How much storage does it use? A: Roughly 20 GB per month of active use, depending on screen complexity and capture settings.

Q: Can I use it without an AI agent? A: Yes. The local web UI at localhost:3030 lets you browse and search your history directly.

Q: What is the license for commercial use? A: Screenpipe uses a commercial license. Personal and non-commercial use is free. Organizations must obtain a commercial license for production use.

Q: Does it work on Linux? A: Yes, but requires building from source. Native installers are available for macOS and Windows.

Conclusion

Screenpipe solves the context problem for AI agents — instead of starting every conversation from scratch, it lets agents see what you’ve been working on. The local-first architecture means your screen history never leaves your machine, which addresses the privacy concerns that make Rewind-style tools non-starters for security-conscious teams.

The MCP integration makes it easy to drop into an existing workflow: npx screenpipe setup wires it into Claude, and you can start asking questions about your work history immediately. The PII filtering model is a thoughtful addition for anyone worried about sensitive data ending up in an AI prompt.

If you want AI agents that actually understand your work context, Screenpipe is currently the most actively developed open option — 20K GitHub stars and YC S26 backing suggest it will be around for a while.

HN discussion: news.ycombinator.com/item?id=49024620