Louis 推出了 Screenpipe，这是一款 YC S26 初创项目，旨在为 AI 智能体提供可搜索的工作记忆 [1]。该应用在本地录制屏幕和音频 [1]，通过监听操作系统事件、配对截图和系统可访问性树来构建本地 SQLite 数据库 [1]。
Screenpipe 采用 Rust、MLX 和 Onnx 等技术栈开发 [1]。其 AI PII 模型在本地运行，占用资源极少，仅需不到 1% 的 CPU 和少于 400 MB 的 RAM [1]。该工具支持多种 AI 模型进行任务自动化、信息检索和个人知识库维护 [1]。
在许可方面，Screenpipe 对个人非商业、非营利、教育和研究使用免费，商业使用需要付费许可证 [1]。许可证改变前的版本仍可在 MIT 许可下获得 [1]。应用提供桌面版和 CLI 两种形式，支持 macOS、Windows 和实验性 Linux 支持 [1]。源代码可在 GitHub 获取 [1]。
Louis has launched Screenpipe, a Y Combinator S26 project designed to record screen and audio locally, creating a searchable work memory for AI agents [1]. The application constructs a local SQLite database by monitoring operating system events and pairing screenshots with system accessibility trees, enabling AI models to automate tasks, retrieve information, and maintain personal knowledge bases [1].
The platform is built using Rust, MLX, and Onnx technologies [1]. Its AI PII model runs locally while consuming less than 1% CPU and under 400 MB of RAM [1].
Screenpipe follows a freemium model: personal, non-commercial, non-profit, educational, and research uses are free, while commercial applications require a paid license [1]. Versions released prior to license changes remain available under the MIT license [1]. The source code is publicly accessible at https://github.com/screenpipe/screenpipe [1].
The application is available as both a desktop application and command-line interface, with support for macOS, Windows, and experimental Linux support [1].