All tools
HarnessCommunity63k

Ruflo

Updated Jul 5, 2026

An agent meta-harness (formerly claude-flow) that adds 100+ specialized agents, coordinated swarms, self-learning HNSW memory, agent federation across machines, and 35 Claude Code plugins to Claude Code and Codex. Install via `npx ruflo init` for the full loop or individual plugins via the Claude Code marketplace.

View on GitHub

What it does

  • npx ruflo@latest init wizard

    Interactive setup wizard that installs the full Ruflo loop: 98 agents, 60+ commands, MCP server, hooks, and background workers.

  • claude mcp add ruflo -- npx ruflo@latest mcp start

    Register Ruflo as an MCP server so memory_store, swarm_init, agent_spawn, and other tools are callable from Claude.

  • /plugin install ruflo-swarm@ruflo

    Install the swarm plugin to coordinate multiple agents as a team with hierarchical, mesh, or adaptive topologies.

  • /plugin install ruflo-sparc@ruflo

    Install the SPARC methodology plugin: Specification, Pseudocode, Architecture, Refinement, Completion — a five-phase structured agentic development workflow.

  • ruflo federation init

    Initialize zero-trust agent federation with mTLS + ed25519 keypair for cross-machine agent collaboration.

Files (1)

OVERVIEW.mdprimary · markdown · 2.2 KB
## What it does

Ruflo (formerly claude-flow) is an agent meta-harness — the execution layer around Claude Code and Codex that adds 100+ specialized agents, coordinated swarms, self-learning memory, agent federation across machines, and enterprise security guardrails. After `npx ruflo init`, you use Claude Code normally — hooks automatically route tasks, learn from successful patterns, and coordinate agents in the background.

**Core capabilities:**

- **100+ agents:** coders, testers, reviewers, architects, security auditors organized by domain.
- **Swarm coordination:** Hierarchical, mesh, and adaptive topologies with consensus protocols (Raft, Byzantine, Gossip).
- **HNSW vector memory:** AgentDB — persistent memory across sessions, measured ~1.9x faster at N=20k vs brute force. Agents recall past solutions via vector search.
- **Self-learning:** SONA neural patterns and ReasoningBank learn from every task; intelligent routing achieves 89% accuracy.
- **Agent federation:** Zero-trust cross-machine collaboration. Agents on different machines or orgs discover each other via mTLS + ed25519, exchange work with PII stripped before transmission, and build behavioral trust scores over time.
- **Background workers:** 12 auto-triggered workers (audit, optimize, testgaps, etc.) run without user intervention.

**35 Claude Code plugins include:** ruflo-core, ruflo-swarm, ruflo-sparc (SPARC methodology), ruflo-rag-memory, ruflo-intelligence, ruflo-security-audit, ruflo-aidefence, ruflo-browser, ruflo-testgen, ruflo-ddd, ruflo-adr, ruflo-cost-tracker, ruflo-neural-trader, and more.

## Install

**Two paths:**

**Path A — Claude Code plugins (slash commands only, no MCP):**

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

**Path B — Full CLI install (MCP server, hooks, daemon, all 98 agents):**

```bash
npx ruflo@latest init wizard
```

Add as MCP server:

```bash
claude mcp add ruflo -- npx ruflo@latest mcp start
```

## When to use

Large parallel workloads where sequential execution is too slow — refactors across many files, documentation at scale, multi-hour autonomous sessions with cross-machine agent collaboration.