ruvnet/ruflo

Pitched as "the leading agent meta-harness for Claude" — multi-agent swarms, coordination, adaptive memory, RAG, native Claude Code / Codex integration. Strong marketing copy and broad claims; audit before integration.

What it is

A TypeScript meta-harness framework for multi-agent AI workflows. README is marketing-heavy with broad claims (swarm intelligence, self-learning, adaptive memory, RAG). Companion site at Cognitum.One. MIT-licensed.

Key features

Multi-agent swarm coordination.
Adaptive memory layer.
RAG integration.
MCP server support.
Native Claude Code + Codex integration claims.
npm-distributed.
MIT-licensed.

Tech stack

TypeScript primary.

When to reach for it

You're cataloguing the agent-harness space and want hands-on exposure.

When not to reach for it

You want vendor-supported tooling — first-party offerings are safer.
The broad marketing claims warrant verification — "swarm intelligence" + "self-learning" + "adaptive memory" in one project's pitch deserve a hands-on look at what's actually implemented.

Maturity signal

57k stars, 7k forks, MIT. The agent-harness space has high marketing-velocity; verify the actual capabilities against the claims before integration.

Alternatives

anthropics/claude-code — first-party.
obra/superpowers, affaan-m/ECC — comparable third-party.
LangChain / LangGraph — code-first.

Wiki: ruvnet/ruflo