Ruflo + Claude Code: agent swarm setup and launch

This tutorial introduces Ruflo as an open source orchestration layer for Claude Code. The presenter describes specialist agents that coordinate work in parallel and retain information across sessions. The walkthrough focuses on terminal setup and a website task using the coding assistant.

Setup includes default, minimal, full and custom presets, followed by swarm topology and concurrency settings. The presenter selects hierarchical mesh coordination, keeps the default concurrency and chooses a hybrid AgentDB and SQLite memory backend. Other memory options shown include AgentDB alone, SQLite alone and in-memory storage. The setup also covers indexing, learning features and embedding model selection.

For the demo, the presenter asks the swarm to build an AI SEO website. The reported status lists an architect, front-end developer, SEO specialist and reviewer, with some agents waiting for others to finish. The transcript shows task initiation and coordination, but does not establish the quality of a completed site.

The presenter says Ruflo uses HNSW vector search for persistent memory and can connect to Ollama and other providers. Those connections are mentioned rather than demonstrated. A web UI is another option, though the walkthrough uses Claude Code. Claims about speed and self-improvement are the speaker's assertions. The closing Q&A says this setup does not work with Kanban and points viewers to separate Hermes training.