Ruflo

Quick Verdict

Ruflo is the MIT-licensed agent meta-harness in the canonical ruvnet/ruflo GitHub repository. As of July 21, 2026, the current stable release is v3.32.9. It is best suited to engineering teams that want to add specialized agents, swarm coordination, persistent memory, MCP tools, model routing, and cross-machine federation around Claude Code or Codex. It is not a model and it is not a turnkey managed coding service.

The project is also in a naming and package transition from Claude Flow to Ruflo. The ruflo, @claude-flow/cli, and claude-flow packages were released together at v3.32.9, and current documentation still contains some claude-flow commands. New deployments should treat the Ruflo repository and ruflo package as canonical, pin a tested version, and avoid assuming the old and new names describe independent products. The breadth is useful for platform teams, but it creates more operational and security work than a conventional coding assistant.

Best For

• Agent platform engineers evaluating swarms, memory, MCP, routing, and federation
• Advanced development teams already using Claude Code or Codex for repository work
• Self-hosting teams able to operate Node.js, databases, containers, model credentials, and network controls
• Research and prototyping groups comparing coordination topologies and memory strategies
• Not ideal for users who only need autocomplete, teams without dependency and permission review, or organizations seeking a simple hosted product with a clear SLA

Key Features

• Two installation surfaces: Claude Code plugins provide commands, skills, and agent definitions, while npx ruflo@latest init enables the fuller MCP, hooks, daemon, memory, and project configuration surface
• Multi-agent orchestration: specialized agents, hierarchical or mesh topologies, routing, parallel work, and reusable workflows
• Persistent memory: AgentDB, vector retrieval, ReasoningBank, SONA, and trajectory storage for cross-session context
• Multiple providers: connections for Claude, OpenAI, Gemini, Cohere, Ollama, and compatible endpoints
• MCP automation: tools for files, terminals, memory, review, and other workflows
• Federation: identity, trust policies, and work exchange between agents on different machines or organizational boundaries
• Operations controls: observability, token cost tracking, budgets, security scanning, and workflow controls that still require local configuration

Use Cases

• Split a complex repository task among planning, implementation, testing, security, and documentation agents
• Preserve architecture decisions and previous fixes in cross-session project memory
• Automate repeatable test-gap, review, documentation, and dependency workflows
• Compare local and cloud models for quality, latency, and cost
• Research federated agent collaboration in an isolated environment with explicit data controls

Pricing

Ruflo v3.32.9 is MIT-licensed, so the software itself has no license fee. Running it is not cost-free. Cloud model calls may incur Anthropic, OpenAI, Google, OpenRouter, or other provider charges. Local models consume hardware and electricity. A full deployment may also require hosting, MongoDB or other storage, vector indexes, logs, backups, bandwidth, and engineering support. The lighter Claude Code plugin path has a smaller footprint than the full CLI initialization.

Version 3.32.9 was released on July 20, 2026 and includes statusline and memory/SQLite integrity fixes. Because transitional Claude Flow packages remain in the ecosystem, production scripts should pin ruflo@3.32.9 or a later internally validated release rather than relying permanently on @latest.

Pros

• MIT-licensed source with inspectable code and deployment options
• Combines agents, MCP, memory, workflows, routing, and federation in one framework
• Offers both a lighter plugin path and a full CLI path
• Broad coverage for swarm experiments and persistent project context
• Active release cadence, with recent SQLite integrity fixes in v3.32.9

Cons

• Fast-moving features, package names, and documentation increase migration risk
• Multi-agent execution can multiply token, compute, database, and hosting costs
• Full initialization writes project configuration and enables high-privilege MCP, hooks, and daemon surfaces
• Persistent memory and federation expand the impact of prompt injection, data leakage, and poisoned long-term state
• CVE-2026-59726 affected Ruflo versions below 3.16.3: the default Docker Compose MCP bridge could allow unauthenticated remote code execution; previously exposed systems need key rotation and AgentDB/MongoDB inspection, not only an upgrade
• Vendor security and benchmark claims still require validation against the exact version and deployment configuration

Alternatives

If a single coding agent is sufficient, Claude Code or Cursor will usually be easier to operate. Ruflo makes sense when shared memory, parallel roles, or federation produces a measurable gain that outweighs the added system complexity.

FAQ

What is the relationship between Ruflo and Claude Flow?

Claude Flow is transitioning to Ruflo. At v3.32.9, ruflo, @claude-flow/cli, and claude-flow packages still coexist, and some current examples retain old commands. Use the Ruflo repository and release notes as the canonical reference for new work.

Is Ruflo completely free?

The MIT-licensed software is free, but model APIs, Claude Code access, local hardware, hosting, databases, logs, bandwidth, and maintenance can all cost money.

Is v3.32.9 automatically production-ready?

No. Pin it, test it in isolation, review dependencies and permissions, validate database behavior and rollback, and keep human approval around high-impact actions.

Is Ruflo federation secure by default?

The project documents mTLS, signatures, trust levels, PII controls, and audit trails. Those are design and implementation features, not automatic proof of compliance. Security depends on version, configuration, peers, and network exposure.

Which vulnerability should operators know about?

CVE-2026-59726 was a critical unauthenticated RCE in the default Docker Compose MCP bridge for versions below 3.16.3. Operators of previously exposed instances should patch, firewall ports, rotate provider keys, and inspect AgentDB and MongoDB for tampering.

How does the plugin path differ from the full CLI?

Plugins mainly add commands, skills, and agent definitions. Full CLI initialization adds project files, MCP, hooks, daemon, memory, and a much larger execution surface, so it also needs stronger controls.

Bottom Line

Ruflo v3.32.9 is a broad, open-source orchestration layer for teams with the engineering and security capacity to manage a high-privilege agent system. Its integration of swarms, memory, MCP, routing, and federation is the main advantage; package transition, operating costs, and expanded attack surface are the tradeoffs. Adopt it by pinning versions, testing in an isolated repository, minimizing tool permissions, tracking provider costs, and threat-modeling federation and persistent memory before production use.