Ruflo is an agent meta-harness built in TypeScript that provides the execution layer and coordination infrastructure for Claude Code and Codex. Rather than being a standalone AI model, Ruflo functions as the harness that equips Claude with tools, memory systems, execution loops, sandboxes, and control mechanisms so that AI agents can perform actual work. The project is maintained primarily by ruvnet, with secondary contributions from sparkling and coygeek tracked across 3223, 132, and 79 events respectively.
The repository offers two distinct installation paths. The lighter Claude Code Plugin path provides slash commands and agent definitions without registering the full Ruflo MCP server, making it suitable for users wanting to try individual plugins. The full CLI installation via npx ruflo init provides the complete system including 98 agents, 60+ commands, 30 skills, an MCP server, hooks system, and daemon functionality. The full installation creates configuration files in the workspace including .claude/, .claude-flow/, CLAUDE.md, and various helpers and settings files.
Ruflo's architecture is organized around 35 specialized plugins grouped into six functional categories. Core orchestration plugins handle server operations, agent coordination through swarms, autonomous execution loops, background task scheduling, workflow templates, and federated communication across machines. Memory and knowledge plugins provide vector database functionality, hybrid search with graph traversal, session persistence, GPU-accelerated search through ruvector integration, and knowledge graph construction. Intelligence and learning plugins enable agents to improve from past successes, perform sublinear graph reasoning with PageRank algorithms, manage dynamic agent behavior, run local language models with smart routing, and break goals into trackable plans.
Additional plugin categories address code quality through test generation and browser automation, security via vulnerability scanning and prompt injection defense, and architectural methodology through architecture decision records, domain-driven design scaffolding, guided development phases, meta-harness grading, and competitive agent tournaments. The system emphasizes self-organization where agents automatically coordinate into swarms, learn from successful patterns, and maintain memory across sessions without requiring users to master hundreds of tools or commands.
GitGenius tracking shows the repository has 421 subscribers as of the most recent check, with 725 open issues. The project maintains a median issue and pull request response latency of 2.2 hours with a mean of 94.2 hours across 1285 tracked items. Enhancement requests represent the most active issue category with 91 items, followed by bug reports with 43 and verification tasks with 42. The ecosystem demonstrates significant adoption with 8.1 million downloads and 106,000 git clones in a 14-day period. The project connects to related repositories including anthropics/claude-code, opendatahub-io/notebooks, and openclaw/openclaw through overlapping contributors. Ruflo is distributed via npm and available under MIT license, with a web-based UI beta accessible at flo.ruv.io and powered by the Cognitum.One agentic architecture running a Rust-based AI engine.
