Portable AI agent desktop for Windows — 100 tools, GUI, local models via LM Studio, TTS, Music, ComfyUI, workflows, tool maker. No install. No Docker. No admin rights.
Built on NousResearch/hermes-agent (MIT License) with extensive customization for non-technical users.
- Dark-themed Tkinter interface with chat, sidebar, and session management
- Image attachment with thumbnails (vision model support)
- Guided mode — works even without an AI model connected
- API key setup wizard with individual service configuration
- Permissions panel with granular control over file, network, and system access
| Toolset | Tools | What It Does |
|---|---|---|
| LM Studio | 10 | Load/unload models, search HuggingFace, tokenize, embed, direct chat |
| Music | 7 | Generate music, manage models, GPU workers, output library |
| TTS | 7 | Text-to-speech, 10 voice models, voice cloning, job management |
| ComfyUI | 7 | Image generation, instance management, model/node browsing |
| Workflows | 6 | Create, run, schedule, and manage multi-step automation pipelines |
| Tool Maker | 3 | Dynamically create API wrapper or Python handler tools at runtime |
| Serper | 1 | Google-quality search via Serper.dev API |
| Guide | 1 | Searchable built-in user manual |
| GPU | 1 | NVIDIA GPU status (memory, temp, utilization) |
| Model Switcher | 1 | Switch between cloud and local AI models |
| Hermes Update | 2 | Pull upstream updates + auto-reinject custom tools |
Plus all built-in hermes-agent tools: web search, file operations, browser automation, code execution, delegation, memory, skills, messaging, Home Assistant, and more.
Three portable AI generation servers from rookiemann:
| Extension | Port | Models | GPU |
|---|---|---|---|
| TTS Server | 8200 | Kokoro, XTTS, Dia, Bark, Fish, + 5 more | 4 GB+ |
| Music Server | 9150 | MusicGen, Stable Audio, ACE-Step, Riffusion | 4 GB+ |
| ComfyUI | 5000 | SD 1.5, SDXL, Flux, 100+ registry models | 6 GB+ |
Each extension auto-installs on first use. No system dependencies.
Chain tool calls into automated pipelines with data flow, conditions, loops, parallel execution, error handling, and cron scheduling.
Create new tools at runtime — wrap any REST API or write custom Python handlers. Tools persist across sessions and reload automatically.
No API key? No problem. The chat works offline using a built-in 1,054-line user guide. New users get step-by-step guidance to set up their first AI model.
install.batDownloads embedded Python 3.13, all dependencies, LM Studio SDK, and Node.js tools. No admin rights needed.
hermes.bat :: CLI mode
hermes_gui.bat :: GUI modeCloud (2 minutes, free):
- File > API Key Setup > OpenRouter
- Sign up at openrouter.ai (free, no credit card)
- Paste your API key
- Start chatting
Local (needs NVIDIA GPU):
- Download LM Studio
- Download a model, start the server
- Tools > LM Studio in the GUI
- Load model, click "Use for Chat"
- Windows 10/11
- Internet connection (for cloud AI) or NVIDIA GPU 8GB+ (for local AI)
- No admin rights, no system Python, no Docker
A searchable user guide is built into the agent — ask it anything or use the search_guide tool. The PDF manual is included in every release.
Key topics: getting started, API setup, the interface, permissions, LM Studio local models, extensions (TTS/Music/ComfyUI), all 100 tools, custom tool creation, workflows, and a glossary of AI terms.
User
|
v
GUI (Tkinter) / CLI
|
v
Agent Bridge (threading, sessions)
|
v
AIAgent (run_agent.py)
|
+-- Tool Registry (100 tools)
| +-- LM Studio tools (SDK + HTTP)
| +-- Extension tools (Music, TTS, ComfyUI)
| +-- Workflow engine
| +-- Tool maker (dynamic creation)
| +-- Serper, GPU, Guide, etc.
| +-- Custom tools (user-created)
|
+-- LLM Provider
+-- OpenRouter (cloud)
+-- LM Studio (local, GPU)
+-- Any OpenAI-compatible endpoint
- Base framework: NousResearch/hermes-agent (MIT License)
- Extension modules: rookiemann — portable-tts-server, portable-music-server, comfyui-portable-installer
- Custom tools, GUI, and integrations: Built with Claude Code
MIT License — see LICENSE for details.
Original framework copyright (c) 2025 Nous Research.