Awesome AI Agents › AI Agent Orchestration

DmitrySolana/ralph-claude-code

⭐ 747 Shell repository created 2026-01-16

This repository is a public template for building an Agent Control Room, a sidecar folder that documents and governs a fleet of Hermes agents running on a VPS. The Control Room is deliberately not an agent itself. It is the system map, operating manual, registry, runbook library and recovery notebook for the agents that are actually running, and the stated method is to set up that control plane first and only then plug agents into it. The template supplies a control-plane folder structure, templates for per-agent runbooks, Docker notes, secret and port maps and backup procedures, plus bundled setup and operations skills that an agent can invoke to build or maintain the system. Its architecture is described in levels that mirror how a personal agent setup tends to grow: start with one agent, add direct specialists once roles become clear, add an orchestrator when a single front door is wanted, and automate only after the manual arrangement works. Specialists in the worked example cover personal assistance, SEO, development, marketing and operations. Delegation between the orchestrator and specialists runs through an Agent Task Bus, a filesystem handoff area with inbox, working, outbox and archive stages, so requests and results are auditable artifacts rather than ephemeral chat. The operator keeps three access paths at all times: editing the Control Room directly, talking to any specialist directly, or going through the orchestrator, which routes tasks and synthesizes a final answer. Content is mostly Markdown documentation and shell setup scripts, with Mermaid diagrams of the system shape. It targets people self-hosting several Claude Code style agents who want operational discipline rather than a new framework.

https://github.com/DmitrySolana/ralph-claude-code

agent-operationsorchestrationtask-busrunbookstemplateclaude-codeself-hostingmulti-agent