What Is Ruflo?
Ruflo, by ruvnet, is the reference orchestration platform for the rest of this course. It is interesting because it is comprehensive — almost every multi-agent concept the course discusses is implemented, named, and configurable in ruflo. The downside of the comprehensiveness is that ruflo's documentation is dense; this concept gives you the architectural tour so the rest of the course can refer to specific ruflo subsystems without re-explaining each one.
Ruflo's elevator pitch: it adds a "nervous system" to Claude Code. Agents self-organize into swarms, learn from past tasks, and remember across sessions. The user keeps writing code in Claude Code; ruflo handles the coordination.
How It's Architected
The high-level diagram:
Concretely, that decomposes into:
ruflo-core— the orchestration entry point, plus health checks, plugin discovery, foundation utilities.- Router — directs incoming tasks to appropriate agents (claimed 89% routing accuracy).
- Swarm Coordinator — manages a swarm, applying a topology and consensus rule.
- Memory subsystem —
AgentDB(HNSW vector DB) + knowledge graphs. - Learning engine —
SONA(self-learning patterns),ReasoningBank, trajectory analysis. - Federation layer — cross-machine collaboration with zero-trust security (mTLS + ed25519 + behavioral trust scoring).
- Worker pool — 12 background workers (audit, optimize, testgaps, dependency-scan, …).
- Policy / security pipeline — input validation, CVE scanning, PII gating across 14 PII types, prompt-injection defense (
AIDefence).
Why It Matters
Two reasons. First, ruflo is the most concrete reference for harness-level multi-agent patterns. Pointing at ruflo-loop-workers or ruflo-aidefence is shorter than hand-waving about hypothetical implementations. Second, ruflo is the pragmatic choice when a team needs orchestration today — it is open-source, packageable as Claude Code plugins, and has a marketplace of pre-built agents and workflows.
Key Technical Details
- Plugin model: 32 native plugins (e.g.,
ruflo-core,ruflo-swarm,ruflo-autopilot,ruflo-loop-workers,ruflo-workflows,ruflo-federation,ruflo-rag-memory,ruflo-knowledge-graph,ruflo-intelligence,ruflo-goals,ruflo-aidefence,ruflo-sparc). - Topologies: Queen-led (one orchestrator, many workers), mesh, hive mind, adaptive (runtime-switched).
- Consensus: Raft, Byzantine, gossip — selectable per swarm.
- Federation trust pipeline: PII strip → ed25519 sign → mTLS verify → behavioral trust evaluate → privilege adjust.
- Background workers: 12 auto-triggered, each with its own goal and budget.
- GOAP planner: A* search over annotated actions, with adaptive replanning.
- Reported metrics (May 2026): 84.8% solve rate on SWE-bench, 75% claimed cost savings vs raw Claude Code.
- Methodology plugins:
ruflo-sparc(Specification, Pseudocode, Architecture, Refinement, Code),ruflo-adr,ruflo-ddd— encode whole engineering methodologies as agent workflows.
How Harnesses & Frameworks Implement This
| Concern | Ruflo's choice | Closest equivalent in another harness |
|---|---|---|
| Topology | Queen-led default, multi-mode | Claude Code: supervisor sub-agents only |
| Memory | AgentDB + ReasoningBank + SONA | LangGraph: checkpointers; Claude Code: CLAUDE |
| Federation | mTLS + ed25519 zero-trust | None standard; mostly DIY |
| Learning | Trajectory + LoRA adapters | None standard |
| Background workers | 12 first-class workers | Claude Code: DIY via SessionStart hooks |
Connections to Other Concepts
- What Is an AI Harness? — The category. Ruflo is positioned as a harness atop a harness.
- queen-led-hierarchy, hive-mind-pattern, adaptive-topology-switching — Ruflo's signature topologies.
- reasoning-bank, sona-self-learning-neural-patterns, agentdb-and-vector-stores-in-harnesses — Memory subsystems.
- cross-machine-agent-federation, mtls-and-ed25519-for-agent-trust, behavioral-trust-scoring — Federation primitives.
- methodology-as-plugin-sparc, methodology-as-plugin-adr-and-ddd — Methodology plugins.
Further Reading
- ruvnet, ruflo on GitHub — README + USERGUIDE are the single best entry points.
- ruvnet, ruflo CHANGELOG — A real-time view of which subsystems are most active.
- "Ruflo: Claude Multi-Agent Orchestration Platform Guide" (AIToolly, 2026) — Useful third-party overview.