ruvos
rUvOS — the agentic operating system. A single static Rust binary that runs as an MCP server, giving Claude Code persistent memory, resumable signed sessions, multi-agent orchestration, a knowledge graph, safety guardrails, and a signed audit trail. Zero Node.js, zero SQLite, zero daemon.
rUvOS is a single Rust binary that gives AI coding assistants like Claude Code persistent memory, resumable sessions, multi-agent coordination, and a tamper-evident audit log, no server, database, or cloud account required.
rUvOS is a single program, compiled to a self-contained binary in Rust, that gives AI coding assistants a persistent memory and the ability to coordinate teams of specialized agents. By default, tools like Claude Code forget everything between sessions, work alone, and leave no record of what they did or why. rUvOS addresses all three of those gaps.
The program connects through the Model Context Protocol, a standard interface that AI coding tools use to discover and call external capabilities. Once you install rUvOS and restart your assistant, it gains access to 50 additional tools covering memory, session management, agent coordination, safety checks, and an audit log. You do not call these tools directly: you describe what you want in plain English, and the assistant decides which tools to use.
Memory works by storing facts and decisions as vector embeddings, which allows the system to retrieve relevant past context based on meaning rather than exact word matches. Sessions are saved as signed files on your disk, so you can resume a work context days later from exactly where you left off. You can also branch a session before attempting something risky, so you have a fallback if it goes wrong.
The multi-agent side lets your assistant spin up teams of specialist agents, each with a defined role such as coder, tester, or security reviewer. A planner figures out the sequence of steps for a given goal, and if a step fails the pipeline can retry or stop cleanly. Separate instances of Claude Code running in different terminal windows can also discover each other and pass messages using plain files on disk, without any shared server or database.
Every action passes through a safety check before it runs, and a signed audit log records what happened. The log is tamper-evident, meaning any modification after the fact would be detectable.
The project acknowledges it consolidates and re-implements an ecosystem of tools originally created by Reuven Cohen. It requires no Node.js, no SQLite, no cloud account, and no background service. The setup script builds the binary, installs it, and registers it with Claude Code in one step.
Where it fits
- Give Claude Code a persistent memory so it remembers decisions and context from previous sessions without re-explaining your project each time
- Coordinate a team of specialist AI agents, coder, tester, security reviewer, to tackle a complex feature from a single plain-English goal
- Branch your AI session before a risky refactor so you can roll back cleanly if the automated changes go wrong