- Category
- AI Agents
- Rank
- No. 32Tools index
- Type
- APP
- Builder
- ruvnet
- GitHub
- 71.7k stars
- Latest release
- v3.38.23
- Date
About
Ruflo (formerly Claude Flow) is multi-agent orchestration for Claude Code, coordinating 100+ specialized agents across machines, teams, and trust boundaries with swarms, self-learning memory, federated comms, and enterprise security.
What it does
Ruflo wraps coding assistants with routing, persistent memory, background hooks, reusable workflows, and coordinated specialist workers. You can start with lightweight Claude Code plugins or install the full runtime, which adds terminal commands, an MCP server, workspace configuration, hooks, and a daemon.
Why it's ranked here
Ruflo is compelling when one coding agent is no longer enough. Its broad command, plugin, memory, security, and federation surfaces address real coordination work. The repository also documents unresolved integration coverage, Windows daemon behavior, branding inconsistencies, and unvalidated load scenarios, so adoption should include careful testing.
What's good
The two installation modes make commitment explicit. Plugins can add focused commands and agent definitions without writing workspace files, while the full setup supplies routing, hooks, memory, MCP access, and diagnostics. Federation adds signed identities, trust levels, spending limits, circuit breaking, PII checks, and an audit trail.
Tradeoffs
The full setup writes configuration and helper state into the project, which increases operational footprint. The lighter plugin route omits hooks and most MCP coverage unless the core plugin is installed. Documentation contains differing capability counts across snapshots, while the package depends on several alpha components and optional native or database modules.
How to use it well
Use Ruflo for complex engineering work that benefits from task routing, specialist agents, shared memory, repeatable workflows, or controlled collaboration between machines. Begin with focused plugins, then adopt the full runtime when hooks and MCP coordination justify workspace changes. Federation does not replace human chat, identity vetting, NAT traversal, or general-purpose RPC.
Technical notes+
package.json defines a public ESM npm package, requires Node.js 20 or newer, exposes a CLI binary, runs Vitest, and bundles the Codex, federation, and security packages. Cargo.toml says TypeScript is primary while registering separate Rust federation and AGNTCY crates. docs/index.md describes the Claude Code plugin marketplace and MCP surface. docs/STATUS.md documents the CLI, MCP, plugin, WASM, verification, and test surfaces. docs/QUALITY-SWEEP.md records remaining dead-export work and deferred profiling. docs/IMPROVEMENT-ROADMAP.md identifies skipped integration tests, Windows daemon persistence, branding leakage, and missing real-model validation. docs/federation/README.md specifies signed peer manifests, trust gates, budgets, circuit breaking, WSS transport, and optional WireGuard projection.
Observed
- License
- MIT
- Primary language
- TypeScript, with Rust components for federation and related workspace crates
- Packaging
- Public ESM npm package with an npx setup path and Claude Code plugin marketplace
- Interfaces
- CLI, MCP server, Claude Code plugins, and bundled Codex integration
- Runtime
- Node.js 20 or newer
- Testing
- Vitest test script, dedicated security test script, and a declared tests directory
Read from README.md, Cargo.toml, package.json, docs/index.md, docs/STATUS.md, docs/USERGUIDE.md, docs/QUALITY-SWEEP.md, docs/IMPROVEMENT-ROADMAP.md, docs/TEAM-GATEWAY-CHECKLIST.md, docs/metaharness-user-guide.md, docs/darwin/PLAN.md, docs/darwin-core/PLAN.md, docs/federation/README.md.
What it can do
Orchestrate multiple AI agents across distributed systems
AI agents and task requirements → Coordinated agent execution across machines and teams
Enable AI agents to self-organize into collaborative swarms
Individual AI agents and task parameters → Self-organized agent swarms with coordinated behavior
Store and retrieve agent learning patterns across sessions
Agent task execution data and outcomes → Persistent memory and learned patterns for future tasks
Route tasks automatically to appropriate AI agents
User tasks and available agent capabilities → Optimal task distribution to specialized agents
Enable secure federated communication between agents on different machines
Cross-machine agent communication requests → Secure data exchange without data leakage
Run AI agents autonomously in continuous loops
Agent configuration and autonomy parameters → Self-executing agent workflows without manual intervention
Schedule and execute background tasks on timers
Task definitions and timing schedules → Automated execution of scheduled background processes
Create reusable workflows from successful agent patterns
Successful agent execution patterns and outcomes → Templated workflows for repeated use
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.
