Louis Beaumont (@louis030195) presented Screenpipe to Y Combinator on July 14th as a memory layer that records computer activity and gives AI agents a searchable account of how their user works.
https://x.com/screenpipe/status/2077045452579778664 The July 14th launch post described Screenpipe as software that converts screen and audio recordings into searchable memories, standard operating procedures and agents. The pitch turns passive desktop recording into infrastructure for agents that otherwise depend on users manually supplying documents, meeting notes and prompts.
Beaumont created Screenpipe in 2024 and remains its sole founder, according to Screenpipe's company page. He wrote in a LinkedIn post that he built the first version during a weekend in Thailand while running a defense AI startup. Beaumont said his earlier work included French intelligence and that he had spent years trying to give language models access to personal context before concluding that the richest record was the activity already visible on a user's screen.
That thesis now sits behind a six-person San Francisco company in Y Combinator's Summer 2026 batch. Screenpipe disclosed its acceptance into the batch on May 20th, nearly two months before its public YC launch. YC lists Gustaf Alstromer as Screenpipe's primary partner.
How Screenpipe records work
Screenpipe runs in the background on macOS, Windows and Linux. Its documentation says the software reads structured text through operating-system accessibility APIs, falls back to optical character recognition when that data is unavailable and transcribes computer and microphone audio. Screen captures, extracted text and transcripts are stored in a local SQLite database by default.
Users can search that history through Screenpipe's desktop interface or local API. An MCP server lets compatible assistants query the archive, while Screenpipe's "pipes" are scheduled agents defined in markdown files. A pipe can summarize a workday, draft meeting notes, update another application or identify a repeated workflow that could be automated.
The distinction matters for Screenpipe's enterprise pitch. A searchable personal timeline helps an individual recover a lost tab or meeting commitment. Deploying the recorder across employee computers could let an organization reconstruct processes that were never documented in a wiki or CRM, then convert those patterns into SOPs and automated tasks.
That approach also creates an unusually broad data collection surface. Screenpipe says users can exclude particular applications, windows and URLs, and can keep transcription and AI processing on-device. The GitHub repository says product analytics through PostHog and crash reporting through Sentry are enabled while telemetry is active. Cloud transcription, hosted AI and synchronization also require selected data to leave the device. Users seeking a fully local setup must disable telemetry and avoid those cloud services.
The open-source claim has a commercial limit
Screenpipe's launch post called the project open source and cited more than 20,000 GitHub stars, 1,900 forks and 130 contributors. The public repository had approximately 20,400 stars and 2,000 forks on July 23rd, supporting the first two claims.
The open-source description is less precise. On June 9th, Screenpipe replaced its MIT license with the Screenpipe Commercial License. The current license file allows personal, nonprofit, educational and research use, plus a seven-day organizational evaluation. Commercial or production use of the source requires a paid license regardless of a company's size, revenue or funding.
Screenpipe itself acknowledged that the new terms are source-available and are not open source under the strict definition. Earlier versions released under MIT remain available under that license, while current and future code falls under the commercial restrictions.
The licensing change exposes the business model behind the YC launch. Screenpipe used a public codebase to build developer distribution and an auditable privacy story, then restricted commercial deployment as it moved toward team sales. Its repository currently lists a Standard subscription at $25 per month, Pro at $50 per seat per month and Enterprise at $150 per seat per month.
Screenpipe is betting that agent developers will pay for a persistent stream of work context rather than assemble that context from individual SaaS integrations. Beaumont's harder task is earning permission to record the screen, microphone and application activity that make the product useful. Local storage, inspectable code and granular exclusions address that concern, while the commercial license gives Screenpipe a way to charge organizations that deploy the recorder at scale.