Hermes Agent Desktop

The Desktop Client That Turns Hermes Agent Into a Full AI Team

Features • What's New • Screenshots • Quick Start • Configuration • Architecture • License

Not just a GUI wrapper — this is a ground-up rebuild of the Hermes Agent experience. We added a complete visual multi-agent collaboration system and an integrated Skill Store that the original CLI version doesn't have.

Hermes Agent by Nous Research is already the most capable open-source AI agent. This desktop client takes it further — replacing the single-agent terminal with a 20-person AI team led by a Project Manager who automatically understands your requirements, decomposes complex tasks, delegates to the right specialists, and delivers integrated results. No prompt engineering required.

Built with a clean Apple-inspired design, zero Electron bloat, and works with any OpenAI-compatible LLM provider (DashScope, DeepSeek, OpenAI, Anthropic, OpenRouter, and more).

What's New vs Original Hermes Agent

Why Multi-Agent Matters

A single AI agent can write code — but building a real product needs a team. This client gives you:

• A Project Manager who breaks "build me an e-commerce platform" into 12 actionable sub-tasks
• A Product Manager who writes the PRD before any code is touched
• A UI Designer who defines the interface before the frontend engineer starts
• 3 Engineers (frontend, backend, full-stack) who write actual code in their domains
• A QA Engineer who catches what the developers missed
• An Architect who ensures the pieces fit together at scale

All orchestrated automatically. You describe what you want; the team delivers.

Features

Multi-Agent Collaboration System

• 20 Built-in AI Agents — Project Manager, Product Manager, UI Designer, Frontend/Backend/Full-stack Engineers, QA, Architect, DevOps, Data/AI Engineer, Security Expert, Operations, Marketing, Business Analyst, Tech Writer, Translator, Legal Counsel, DBA, Creative Director
• Project Manager as Orchestrator — Automatically receives requirements, decomposes tasks, delegates to specialists, tracks progress, and synthesizes deliverables
• Real-time Agent Status — Dashboard showing which agents are active, task progress, and completion stats
• Agent CRUD — Create, edit, and delete custom agents with configurable system prompts, models, and skill tags

Chat Interface

• Streaming SSE Responses — Real-time token-by-token display with typing cursor animation
• Multi-Agent Response Sections — Clearly labeled sections showing which agent contributed what
• Rich Markdown Rendering — Headings, code blocks with syntax highlighting & copy button, tables, blockquotes, lists, inline code
• Tool Call Indicators — Collapsible panels showing agent tool usage, auto-collapsed after completion
• Session Management — Multiple conversations with history, auto-save to localStorage

Integrated Skill Store (New)

The original Hermes Agent requires CLI commands to discover and install skills. We built a visual Skill Store directly into the desktop client:

• 50+ Curated Skills — Handpicked from CocoLoop Skill Hub, covering AI search, browser automation, code execution, data processing, content creation, and more
• One-Click Install — Click "+" to install any skill instantly, with loading animation and toast confirmation
• Smart Search — Real-time fuzzy search across all skill names and descriptions
• Category Tags — Skills organized by type (Search, Agent, Development, Productivity, Security, etc.)
• Direct Store Access — Link to the full CocoLoop marketplace for browsing hundreds more

Desktop Experience

• Native macOS Window — Powered by pywebview with system-native chrome
• Workspace Selector — Native folder picker dialog for setting working directory
• Model Switcher — Quick switch between models (Kimi K2.5, Qwen Plus/Max, DeepSeek V3/R1)
• Settings Panel — Configure API endpoint, key, and model
• Apple-Inspired Design — Clean grey palette, card-based layout, smooth animations

Screenshots

Chat Interface

Clean, distraction-free conversation view with streaming responses and multi-agent section dividers.

Agents Dashboard

Manage 20+ AI specialists with real-time status, skill tags, and tool counts.

Skill Store

Discover and install AI skills from the CocoLoop marketplace.

Settings

Configure your LLM provider and API credentials.

Quick Start

Prerequisites

• macOS (primary), Linux, or WSL2
• Python 3.11+
• Git

Installation

# 1. Clone the repository
git clone https://github.com/Felix-Forever/hermes-agent-desktop.git
cd hermes-agent-desktop

# 2. Clone the Hermes Agent core (dependency)
git clone --depth 1 https://github.com/NousResearch/hermes-agent.git hermes-core

# 3. Create virtual environment
python3.11 -m venv venv
source venv/bin/activate

# 4. Install dependencies
pip install -e "./hermes-core[all,dev]"
pip install pywebview

# 5. Configure your API key
cp .env.example .env
# Edit .env and add your API key

# 6. Launch
python app.py

One-Command Install (with conda)

# Using conda (recommended for macOS)
conda create -n hermes python=3.11 -y
conda activate hermes
git clone https://github.com/Felix-Forever/hermes-agent-desktop.git
cd hermes-agent-desktop
pip install -e "./hermes-core[all,dev]" pywebview
python app.py

Configuration

Environment Variables (.env)

# LLM Provider API Key (required)
DASHSCOPE_API_KEY=your-api-key-here
# Or use OpenAI-compatible keys:
# OPENAI_API_KEY=your-key

# API Base URL
BASE_URL=https://dashscope.aliyuncs.com/compatible-mode/v1

# Default Model
MODEL=kimi-k2.5

Supported LLM Providers

In-App Settings

Click the Settings icon in the sidebar or the gear icon in the input footer to configure:

• API Endpoint — Your LLM provider's base URL
• API Key — Your authentication token
• Model — The model to use for conversations

Architecture

hermes-agent-desktop/
├── app.py              # Backend: aiohttp API server + pywebview launcher
├── index.html          # Frontend: single-file HTML/CSS/JS application
├── .env.example        # Environment variable template
├── README.md           # This file
└── docs/
    └── screenshots/    # App screenshots

How It Works

┌─────────────────────────────────┐
│   pywebview Desktop Window      │
│   ┌───────────────────────────┐ │
│   │   HTML/CSS/JS Frontend    │ │
│   │   • Chat UI (SSE stream)  │ │
│   │   • Agents Dashboard      │ │
│   │   • Skill Store           │ │
│   │   • Orchestrator Logic    │ │
│   └──────────┬────────────────┘ │
└──────────────┼──────────────────┘
               │ HTTP / SSE
               ▼
┌─────────────────────────────────┐
│   Python aiohttp Backend        │
│   • /v1/chat/completions (SSE)  │
│   • /v1/models                  │
│   • /api/choose-folder          │
│   • Creates AIAgent per request │
└──────────────┬──────────────────┘
               │
               ▼
┌─────────────────────────────────┐
│   Hermes AIAgent Core           │
│   • LLM API (OpenAI-compatible) │
│   • Tool Execution (60+ tools)  │
│   • Memory & Skills System      │
│   • Session Persistence (SQLite)│
└─────────────────────────────────┘

Multi-Agent Orchestration Flow

User Input
    │
    ▼
┌─────────────────┐
│  Project Manager │ ← Orchestrator system prompt with all agent definitions
│  (Main Agent)    │
└────────┬────────┘
         │ Task Decomposition
         ▼
┌────────────────────────────────────────┐
│  Sub-tasks assigned to specialists:     │
│  ┌──────┐ ┌──────┐ ┌──────┐ ┌──────┐  │
│  │ UI   │ │Front │ │Back  │ │ QA   │  │
│  │Design│ │ end  │ │ end  │ │ Test │  │
│  └──┬───┘ └──┬───┘ └──┬───┘ └──┬───┘  │
│     │        │        │        │       │
│     ▼        ▼        ▼        ▼       │
│  [Design] [Code]   [API]    [Tests]    │
└────────────────┬───────────────────────┘
                 │
                 ▼
┌─────────────────┐
│  Project Manager │ ← Synthesize all outputs
│  Final Delivery  │
└─────────────────┘

Tech Stack

• Frontend: Vanilla HTML5 + CSS3 + JavaScript (zero dependencies, single-file)
• Backend: Python 3.11 + aiohttp (lightweight async HTTP server)
• Desktop: pywebview (native OS webview, no Electron bloat)
• Agent Core: Hermes Agent by Nous Research
• LLM: Any OpenAI-compatible API provider

Development

# Run in development mode (opens in browser if pywebview not available)
python app.py

# Run frontend only (for UI development)
python -m http.server 8643 --directory .

# Run syntax check on embedded JavaScript
python -c "import re; open('/tmp/c.js','w').write(re.search(r'<script>(.*?)</script>',open('index.html').read(),re.DOTALL).group(1))" && node --check /tmp/c.js

Roadmap

• Real skill installation via Hermes CLI backend
• Agent-to-agent message passing (true parallel execution)
• Workspace file tree browser
• Voice input/output (TTS/STT)
• Plugin system for custom tools
• Dark mode theme
• Windows & Linux native builds
• Auto-update mechanism

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

1. Fork the repository
2. Create your feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

License

MIT License — see LICENSE for details.

Acknowledgments

• Hermes Agent by Nous Research — The powerful AI agent core
• pywebview — Lightweight native desktop webview
• CocoLoop Skill Hub — AI skill marketplace

_{Built with Claude Code by Anthropic}