What is utopia-os?

Most AI coding assistants start every conversation from scratch. They forget what you discussed last week, and nothing they learn carries forward. Utopia OS tries to fix this by wrapping a coding agent in something closer to an operating system, with memory, safety checks, and ways to hand work between multiple agents. The framework adds several layers to a standard coding agent. The memory system has tiers: a core set of always-loaded facts, an on-demand recall layer, and a semantic archive. This means the agent can remember who you are and what decisions were made without re-reading everything each time. There is a single write path for operational state, so things like the current project or remaining budget stay consistent. Security gates force a human confirmation before the agent does anything risky, like installing packages or editing settings. A knowledge pipeline routes the agent's own draft work through a review step before it touches the canonical record. A survival layer tries to recover cleanly from crashes or context limits mid-task. The repo is a skeleton with all personal and business content removed. You get design docs, empty memory templates, security scripts, and reusable skill workflows. Skills include things like a critic, a review gate, and a "sycophancy guard" meant to keep the agent from just agreeing with you. The system prompt skeleton lives in a file called CLAUDE.md that ties everything together. Setup is designed to be done by the agent itself. You point your coding tool at a setup runbook, and it detects which platform you are using, reads the right adapter guide, and wires things up. It is built primarily for Claude Code, but a portable layer runs on any MCP-capable harness. Some platforms are easier to adapt than others. The README is thorough and clearly written, covering architecture, adoption, and compatibility. It is licensed under MIT.