Skip to main content
Back to timeline
ruvnet/rufloSource publication:

Ruflo ships as a Claude Code plugin pack, putting agent swarms, memory and routing inside the editor

Synopsis

Ruflo is released as a Claude Code plugin set whose core, console and mods supply foundation tools, an agent cockpit, and routing with policy enforcement. It packages the execution layer around the model so agents can form swarms, remember across sessions and route tasks inside Claude Code; mods run with the account's permissions and should be reviewed before installing.

AI-generated editorial illustration: ruvnet / ruflo

Interpretation

Ruflo can now be installed as Claude Code plugins, placing agent coordination, memory and task routing inside the everyday coding environment instead of a separate workflow. Previously such coordination had to be assembled separately or reached through another client; now marketplace commands inside Claude Code install the core, console and mods components. The install path requires Claude Code 2.1.287 or later, adding the plugin marketplace, installing ruflo-core, ruflo-console and ruflo-mods, then reloading plugins and running /ruflo.

The execution layer has three parts: core supplies foundation tools, console opens the agent cockpit, and mods add routing and policy enforcement inside Claude Code. Ruflo is described as the execution layer around Claude Code and Codex, adding tools, memory, loops, sandboxes and controls so agents collaborate rather than merely run. The conceptual flow runs from user through CLI or MCP to router, swarm, agents, memory and model providers, with memory feeding experience back into routing; this flow is explicitly labeled conceptual, not a live execution trace.

Perspective

Intended for developers on Claude Code 2.1.287 or later, with three paths: plugin install, MCP connection, and a standalone terminal interface; the console can run without Claude Code via npx ruflo tui, and --plain prints one frame for scripts. The material states no follow-up validation plan.

Mods run with the account's permissions, and the material asks users to review their code before installing; hook support depends on the client, and the console and mods are Claude Code integrations. Local vector retrieval and contrastive updates can run without an LLM but still require embeddings and configured learning modules.