ruvnet/ruflo

Ruflo – an “agent meta‑harness” for Claude Code & Codex

Ruflo is a framework that sits around large‑language‑model (LLM) agents (especially Anthropic’s Claude via the Claude Code plugin) and gives them a full‑featured execution environment. It bundles a CLI, a set of “plugins”, a web UI and a federation layer so that many specialized agents can cooperate, remember past work, and safely talk to agents on other machines.

What you get out of the box

Feature	What it means for you
100 + ready‑made agents	Pre‑built agents for coding, testing, security scanning, documentation, architecture, trading, IoT, etc.
Swarm coordination	Agents are organized into hierarchical or mesh “swarms” that can reach consensus and share work.
Self‑learning loop	Agents log successes/failures; the system updates internal models (SONA neural patterns, ReasoningBank) to improve future behavior.
Vector memory (AgentDB)	Persistent, HNSW‑indexed embeddings let agents recall facts across sessions (e.g., “remember my favorite color”).
Federation	Zero‑trust, authenticated communication lets agents on different machines or organisations collaborate without leaking data.
Multi‑provider routing	Supports Claude, GPT, Gemini, Cohere, Ollama, and any OpenAI‑compatible endpoint; the router picks the best model for each tool call.
Security guardrails	AIDefence, prompt‑injection detection, CVE scanning, path‑traversal protection, and a “MetaHarness” audit that grades the whole agent setup.
Plugin marketplace	35+ plugins (core, memory, intelligence, DevOps, domain‑specific) can be added individually or all together.
Web UI (beta)	A hosted chat UI (flo.ruv.io) that lets you talk to any model while Ruflo runs the same MCP tools the CLI does. It’s also self‑hostable via Docker.
Goal Planner UI	A GOAP‑style planner (goal.ruv.io) that turns plain‑English goals into A*‑searched agent plans and visualises live agent telemetry.

How you start

1. Lite path – Claude Code plugin

   /plugin marketplace add ruvnet/ruflo
   /plugin install ruflo-core@ruflo ruflo-swarm@ruflo …
   

   Adds slash commands and a few agent definitions without creating any files in your repo.

2. Full path – CLI install

   npx ruflo@latest init   # or the interactive wizard
   

   Generates .claude/, .claude-flow/, a CLAUDE.md guide, and spins up an MCP server that the CLI and UI talk to.

3. Web UI – Open https://flo.ruv.io/ (no API key needed) or run the Dockerfile in ruflo/src/ruvocal/ to host it yourself.

Typical workflow

User → Ruflo CLI (or UI) → Router → Swarm → Individual Agents
          ↑                                 │
          └───── Memory ↔ Vector DB ↔ Knowledge Graph
          ↑                                 │
          └───── Learning Loop (updates agent policies)

When you ask Ruflo to “write a new feature and add tests”, the router spawns a coding agent, a test‑generation agent, and a security‑audit agent. The agents run in parallel, store results in the vector memory, and the learning loop records what worked so the next request is faster and cheaper.

Who might use it

• Dev teams that want LLM‑assisted coding, automated testing, and continuous security checks without writing custom prompts.
• Enterprises needing zero‑trust federation so different departments or subsidiaries can share AI work safely.
• Researchers / hobbyists exploring multi‑agent coordination, GOAP planning, or self‑improving local LLMs (via the ruflo-ruvllm plugin).

Where to find more

• Full guide – RuFlo Explained (14‑chapter markdown walkthrough) – docs/ruflo-explained.md.
• Plugin docs – each plugin lives under plugins/ with its own README.
• Benchmarks & audits – see docs/reviews/intelligence-system-audit-2026-05-29.md and the benchmark script in scripts/.
• Source code – the repo contains the CLI (bin/), the MCP server, the web UI (src/ruvocal/), and the plugin implementations.

In short, Ruflo is a comprehensive, extensible platform that turns LLMs into collaborative, memory‑aware, self‑optimising AI agents, with both a command‑line and a web‑based interface.