AI Agent Index

Home / Stacks / Hermes + Codex + Claude Code Stack

WorkflowCurated by AI Agent Indexcomplex to implement

Hermes + Codex + Claude Code Stack

Most people try to force one model to do everything inside a single chat window. This stack splits roles instead, and works significantly better because of it. Hermes Agent acts as the always-on coordinator: it holds persistent memory, manages scheduled automations, handles tool execution on your machine, and connects through Telegram so you can interact from anywhere. OpenAI Codex runs as the main agent brain inside Hermes, handling coordination and tool use that keeps the orchestration layer running. Claude Code operates as a focused coding specialist, called from the terminal by Hermes whenever a bounded coding task needs to be completed. Hermes delegates the coding task, Claude Code writes or fixes the code, and Hermes verifies the result, runs smoke tests, and wires it into the system. The result is a practical personal operations setup: you tell Hermes what you want, it decides whether to handle it directly or delegate to Claude Code, and the system does the work across tools rather than handing you a list of instructions to run yourself. The key mental shift is treating Hermes as the manager, Codex as the agent brain, and Claude Code as the specialist, each doing what it is best at rather than one model attempting everything.

The workflow — 3 agents in sequence

1

Always-on orchestrator and memory layer

Open-source autonomous AI agent by Nous Research with a self-improving learning loop. Runs on your server or desktop app, remembers what it learns. 209.8k GitHub stars.

Visit

Hermes runs persistently on your server. It holds cross-session memory, manages cron-scheduled automations, connects to Telegram for remote access, and decides whether to handle tasks directly or delegate the coding work to Claude Code via terminal.

2
OpenAI Codex logo
OpenAI CodexMCP server

Main LLM brain powering Hermes

OpenAI coding agent for building and shipping software autonomously. Multi-agent parallel execution, Skills, Automations, and Sites. Free limited access; Plus $20/mo; Pro $200/mo; Enterprise custom.

Visit

Codex runs as the primary model inside Hermes, handling the back-and-forth coordination, tool use, and decision-making about which tasks to handle directly and which to route to Claude Code as a specialized coding task.

3

Focused coding specialist

Anthropic's agentic coding CLI that runs in terminal, IDE, Slack, web, and iOS. Pro $17/mo annual, Max $100-$200/mo. 136.4k GitHub stars, MCP-compatible, SOC 2.

Visit

Save this stack

Get this stack plus weekly new stacks and agent updates delivered to your inbox.

Need help implementing this?

Connect with a vetted AI implementation specialist who can set up this stack for your business.

Find an implementation specialist →

Some agents in this stack expose an MCP server

MCP (Model Context Protocol) is an open standard for connecting AI agents to tools and data. The agents that publish a server can be connected into directly by an AI assistant. The others act as clients, meaning they consume external tools rather than exposing their own, so those steps still need conventional integration.

Discussion

No discussion yet

Be the first to ask a question or share your experience with this stack.

← Back to all stacksSubmit your own stack →