wtf is raytsystem-public-os?

raytsystem is a local-first workspace that gives one person or a small team a single place to manage documents, tasks, knowledge, and AI agent workflows, all running on their own machine instead of a cloud account. The interface only listens on the local loopback address, so nothing about it is reachable from outside the computer it runs on, and the project is explicit that it has not reached a stable, production ready release yet. The system provides a command center screen showing the health of the workspace, tasks, agent runs, and safety controls, a documents area for browsing, searching, and editing Markdown files with backlinks and history, and a knowledge universe view for exploring how evidence, tasks, agents, and code relate to each other. It also manages a durable task list with clear states and a permanent history, and lets a user inspect the available agents and skills, including their permissions, without those imported instructions being allowed to run on their own. Several core operations are built in: importing a new source into the knowledge base, querying that knowledge base for an evidence backed answer, checking the knowledge base's integrity, saving a conclusion as a reviewable draft, researching a topic, reviewing a completed run, and running a security review, plus a way to inspect video or other media under a defined policy. Real connections to outside AI providers, external network access, and other effects that reach beyond the local machine are turned off by default, and several integration features are still described as experimental rather than finished. To get started, a user needs macOS or Linux, or Windows through WSL2, along with Python 3.12 through 3.14 and the uv package manager, then clones the repository and runs a setup and start command from a terminal. Node.js is only needed if someone wants to rebuild the web interface themselves, since the built version is already included. The project can also be started by simply asking a supported AI coding agent, such as Claude Code or Codex, to start it in plain language, since those agents read the project's included instruction files. The project is released under the Apache 2.0 license.