The archive · Developer & Business Tools · Product decision · 2024–2026
Screenpipe bets local screen capture is AI's missing memory — 20k stars, YC S26
Louis Beaumont's open-source recorder turns your screen and audio into searchable AI context; 20,000+ GitHub stars, Launch HN, now YC S26.
Screenpipe
What the business is
Screenpipe is a local-first desktop recorder: it captures screen and audio, indexes what you see and hear in a local SQLite database, and exposes that history to AI agents through an API, an MCP server and scheduled 'pipes'.
How it started
Louis Beaumont had maintained a personal 'second brain' since 2020 and built earlier projects (Ava, Embedbase) trying to give language models context, but they only knew what he explicitly saved. He concluded the richest source was the activity already visible on his screen, and built the first version of Screenpipe in a weekend in 2024 while running a defense AI startup. A 2024 Hacker News post about it shaped the product's priorities: consent, local security, CPU usage and signal-to-noise.
What happened
The MIT-licensed repository grew to 20,000+ stars and 130+ contributors. On 2026-06-09 Screenpipe replaced MIT with its own commercial license — source-available, free for personal and nonprofit use, paid for commercial deployment — which drew sharp criticism on HN but set up monetization (Standard $25/month, Pro $50/seat/month, Enterprise $150/seat/month). It joined YC's Summer 2026 batch (disclosed 2026-05-20) and launched publicly on 2026-07-23 as a six-person San Francisco company.
How it ended up
Still live and scaling: public YC S26 launch on 2026-07-23 (88 points, 67 comments on Hacker News), moving from a developer tool toward team and enterprise sales of 'agent memory'.
Background
Screenpipe is a local-first desktop recorder that captures what you see and hear, indexes it in a local SQLite database, and gives AI agents a searchable memory of your actual work. Its bet: the richest context an agent can have is already on your screen — meetings, tabs, fixes, edge cases — so instead of asking users to paste context into prompts, record the work and let agents query it.
Founder Louis Beaumont built the first version in a weekend in 2024 after years of 'second brain' experiments (Ava, Embedbase) that only remembered what he explicitly saved. The open-source MIT project grew to 20,000+ GitHub stars and 130+ contributors. In June 2026 he replaced MIT with the Screenpipe Commercial License — source-available, free for personal use, paid for commercial deployment — and joined YC's Summer 2026 batch.
Screenpipe launched publicly on 2026-07-23: the Launch HN post drew 88 points and 67 comments, with intense debate about privacy, data exfiltration and the license change, while the team pitched team and enterprise plans (Standard $25/month, Pro $50/seat/month, Enterprise $150/seat/month). As of August 2026 the company, six people in San Francisco, is live and selling agents a persistent stream of work context.
What has to be true
- AI agents had solved reading apps and prompts but had no memory of the actual workday; screen and audio capture filled a gap that RAG and MCP could not.
- Local-first design turned a privacy liability into a feature: data stays on-device, inspectable and excludable, positioning Screenpipe as the safe alternative to cloud recorders like Microsoft Recall.
- The license swap shows the open-source growth trap: MIT bought distribution, and a source-available commercial license is how a solo founder converts 20k stars into revenue.
- The 2026-07-23 Launch HN thread (88 points, 67 comments) shows the market is real but contested: developers simultaneously want agent memory and distrust any tool that records everything.
What can be applied
Open source is distribution, not a business model: MIT earned 20k stars and a developer base; a source-available commercial license converts that audience into revenue, at the cost of HN goodwill.
Aftermath
As of 2026-08-16, Screenpipe is live: the core remains free for personal, nonprofit, educational and research use, while paid tiers (Standard $25/month, Pro $50/seat/month, Enterprise $150/seat/month) target teams and enterprises that deploy recording across employee machines. The six-person San Francisco company is in YC's Summer 2026 batch and sells a searchable work timeline that agents query via API, MCP and scheduled pipes — competing with Microsoft Recall on privacy. No funding beyond YC is disclosed; the repo still shows 20,000+ stars.
Sources
- Launch YC: Screenpipe — Record how you work and turn that into agents
- Startup Spotlight: YC's Screenpipe Wants to Record Everything You Do at Work — and Sell It to Your AI Agents
- Launch HN: Screenpipe (YC S26) — Record how you work and turn that into agents
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card