Local-First AI Screen Memory App (screenpipe)

Overview

screenpipe is a desktop app for macOS, Windows and Linux that records what happens on your screen and in your audio, then makes all of it searchable with AI. It reads text through the operating system’s accessibility APIs, falls back to OCR, transcribes meetings and system audio with Whisper running locally, and stores everything in a local SQLite database.

Louis Beaumont created it in 2024 and is the sole founder. The core is source-available. Users can keep everything on their own machine by using local models through Ollama, or connect ChatGPT, Claude or Gemini for search and analysis. It also runs as an MCP server, so AI coding and chat assistants can query a user’s work history directly.

This is a local-first stack. Capture, transcription and storage happen on the user’s device, and the cloud is optional. The business sells convenience and full history on top of that: a free tier, paid personal plans and per-seat business plans.

Realistic income

All figures come from the founder and were not checked by a third party. In February 2026 Louis Beaumont posted on LinkedIn that screenpipe went from $15k to $50k MRR in 4 weeks. In July 2026 he posted that revenue during YC S26 was growing 20-30% week over week. A Dealroom note from September 2026 reports that, at YC S26 Demo Day, he said revenue grew 10x during YC. That note gives no dollar figure.

This is one funded, technically deep founder’s result after years of open-source work. It does not predict what a new builder will earn.

How to build your own

• Don’t try to copy all of screenpipe. Pick one slice of memory, such as meeting-only memory for one profession, or a searchable reading history for researchers.
• Use whisper.cpp for free on-device transcription and SQLite with full-text search for storage. Both are free.
• Run a small local model through Ollama for summaries and Q&A, so your privacy pitch holds up with no API bill. Offer cloud models as an opt-in upgrade.
• Expose an MCP server early. It turns your app into context for AI agents, a strong hook with developers.
• Start with a free tier and a single paid plan. Add team seats once people ask for sync and shared billing.

Risks

• Always-on screen and audio recording raises serious privacy and consent issues, especially for recording other people in meetings. Recording-consent laws vary by place.
• The capture engine is hard engineering across three operating systems, and OS updates can break it.
• Big platforms are competing here (Microsoft Recall, Limitless/Rewind, Granola).
• Revenue figures are founder-reported and not verified by a third party.
• Storage grows fast (the pricing page cites about 30 GB per month of capture), which can put off users with small drives.