AgentHub

What AgentHub does

Hermes runs in a terminal or behind a messaging gateway, and AgentHub adds a web front end that also works on mobile. It reads Hermes's configuration read-only and talks to it through its API, so nothing in the Hermes install has to change. Agents are named and long-lived, with CEO, manager and worker ranks and delegation across three levels. A delegation goes to a strictly lower rank and is traced and visible in the conversation.

Notes in a shared memory are injected into every agent's prompt and linked with double square brackets. A Jarvis page maps the workspace and takes voice commands, including model switches by voice. The Skills and MCP tabs show what Hermes already has configured, and costs are shown in euros. Models can be set per agent or for a single conversation. The code is plain JavaScript with no framework, three dependencies and no build step. Hermes is not mandatory, since any OpenAI-compatible service works.

Key features

• Named, persistent agents with CEO, manager and worker ranks and three-level delegation
• Shared memory notes linked with [[double brackets]] and injected into all agents' prompts
• Skills and MCP tabs that display the catalogue and servers read from Hermes's own config
• Model choice per agent or per conversation, with costs in euros
• Jarvis voice page with a map of the workspace
• Agent tools for web search, page reading, image generation, .xlsx files and .zip archives

When to use it

• Running a standing team of Hermes-backed agents with a clear chain of delegation
• Using Hermes from a phone browser instead of a terminal or chat gateway
• Checking which Hermes skills and MCP servers are active in one view

Who it is for: Hermes Agent users who want a web interface and an organised team of agents instead of a single terminal session.

How it fits with Hermes Agent

Built for Hermes Agent, which it controls through its API while reading its config read-only; any OpenAI-compatible service can be used instead.

Requirements: Node.js and SQLite, plus Hermes Agent or another OpenAI-compatible service

Note: The README is written in French.

FAQ