The interesting bet here isn't another agent framework — it's the claim that the model was never the bottleneck. Ruflo (formerly Claude Flow) treats Claude Code and Codex as the "model" half of an agent and supplies the other half: the harness of tools, memory, loops, and guardrails that lets a coding assistant behave like a coordinated team instead of a single turn-taking session.
What Sets It Apart
- One
npx ruflo initbootstraps the whole loop — roughly 98 agents, 60+ commands, 30 skills, an MCP server, hooks, and a daemon — so the coordination layer is installed rather than hand-wired per project. - A learning loop records which task patterns succeed and reuses them, while a vector-backed AgentDB persists memory across sessions, so agents stop re-solving the same problem from a cold start.
- Federation lets agents on different machines collaborate without shipping raw data between them — the piece that separates it from single-box swarm demos.
- Two deliberately different install paths (lite Claude Code plugins vs. the full CLI loop) let you try one plugin's slash commands before committing the full footprint to your repo.
Great Fit / Look Elsewhere
Great fit if you already work inside Claude Code or Codex and want multi-agent orchestration, persistent memory, and cross-machine coordination without building the plumbing yourself. Look elsewhere if you want a lightweight single-agent setup: the full install writes .claude/, .claude-flow/, hooks, and a daemon into your workspace, and the surface area — hundreds of MCP tools and dozens of commands — is real overhead when you only need a few slash commands.
