Your entire AI stack in one portable folder. No install. No Docker. No admin rights. Just double-click and go.
AgentNate is a local-first AI orchestration platform that brings together large language models, workflow automation, image/video generation, text-to-speech, and music generation into a single unified interface. Everything runs on your machine — no cloud required.
The name comes from n8n (pronounced "n-eight-n" → "Nate"), the open-source workflow engine at the platform's core. AgentNate wraps n8n with an AI layer so agents can build, deploy, execute, and tear down workflows from natural language — no visual editor required.
- 100% Portable -- One folder. Copy it to a USB drive, another PC, a network share — it still works. No system Python, no system Node.js, no PATH changes, no registry entries, no Docker, no WSL, no admin rights. Delete the folder and it's gone.
- One-Click Install -- Run
install.bat. It downloads Python, Node.js, all dependencies, Playwright Chromium, and n8n automatically. Walk away and come back to a working system. - Multi-Provider LLMs -- Load models from llama.cpp (local GGUF), LM Studio, Ollama, vLLM, or OpenRouter. Run multiple models simultaneously across multiple GPUs. Hot-swap between providers mid-conversation.
- Autonomous AI Agents -- Agents with 187+ tools across 19 categories. Web search, code execution, file management, browser automation, workflow building, image generation, and more. Agents plan, reason, delegate to sub-agents, and remember across sessions.
- n8n Workflow Engine -- Full n8n integration with 72 node types. Agents build and deploy complete workflows from natural language. Webhook endpoints, HTTP APIs, LLM chains, conditional logic, scheduling — all without touching the visual editor.
- ComfyUI Creative Engine -- Image generation, video synthesis, upscaling, inpainting, ControlNet, and more. 11 workflow templates with automatic model downloads. Agents craft custom ComfyUI pipelines from text descriptions.
- Text-to-Speech -- XTTS v2, Fish Speech, and Kokoro engines with voice cloning, multilingual support, and real-time streaming. 12 agent tools for programmatic voice generation.
- Music Generation -- Stable Audio Open and ACE-Step engines for text-to-music and lyrics-to-song. Genre detection, CLAP scoring, and batch generation. 12 agent tools.
- Multi-Panel Chat -- Run independent conversations side-by-side, each with its own model, persona, and routing configuration. Compare models, test prompts, or run different tasks concurrently.
- Tool-Level Race Execution -- When an agent calls a creative tool, race N candidates in parallel and pick the best result. Better outputs without manual re-prompting.
- OpenAI-Compatible API -- Drop-in
/v1/chat/completionsendpoint. Works with Continue.dev, Open WebUI, LangChain, n8n's AI nodes, and any OpenAI SDK client.
- Windows 10 or 11 (x64)
- NVIDIA GPU recommended (CUDA-capable, 8GB+ VRAM for local models)
- Internet connection for first-time setup (downloads ~500MB of dependencies)
That's it. No Python, no Node.js, no Git, no Docker, no Visual Studio, no CUDA toolkit.
git clone https://github.com/rookiemann/AgentNate.git
cd AgentNate
install.bat
Or download the ZIP from Releases, extract anywhere, and double-click install.bat.
The installer runs 7 stages automatically:
| Stage | What It Does | Size |
|---|---|---|
| 1 | Downloads Python 3.14.2 embedded from python.org | ~11 MB |
| 2 | Configures ._pth for site-packages, installs pip |
~3 MB |
| 3 | Installs all Python packages from requirements.txt | ~200 MB |
| 4 | Installs Playwright Chromium (browser automation) | ~150 MB |
| 5 | Downloads Node.js 24.12.0 portable from nodejs.org | ~45 MB |
| 6 | Installs n8n workflow engine via npm | ~100 MB |
| 7 | Installs llama-cpp-python CUDA wheel (GPU inference) | ~50 MB |
Every stage is idempotent — if interrupted, just run install.bat again. It skips already-completed stages.
No NVIDIA GPU? Run
install.bat --no-llamato skip the CUDA wheel. You can still use LM Studio, Ollama, or OpenRouter as providers.
launcher.bat :: Opens in your default browser
launcher.bat --server :: API server only (no browser)
launcher.bat --desktop :: Desktop window (PyWebView)
The launcher auto-detects if dependencies are missing and runs the installer if needed. On subsequent launches, it starts instantly.
update.bat :: Update everything
update.bat --python :: Python packages only
update.bat --node :: n8n only
update.bat --cuda :: Reinstall CUDA wheel
AgentNate/
├── install.bat # One-click installer (7 stages)
├── launcher.bat # Start the app (auto-installs if needed)
├── update.bat # Update all components
├── run.py # Python entry point
├── requirements.txt # Python dependencies
│
├── backend/ # FastAPI server (port 8000)
│ ├── server.py # Main server with CORS, WebSocket, static files
│ ├── agent_loop.py # Reusable agent execution loop
│ ├── personas.py # 15 AI personas with tool groups
│ ├── tools/ # 187+ tools in 19 categories
│ │ ├── tool_router.py # Central dispatcher
│ │ ├── agent_tools.py # Sub-agent spawn, routing, memory
│ │ ├── workflow_tools.py# n8n workflow building (19 tools)
│ │ ├── comfyui_tools.py # ComfyUI creative tools (36 tools)
│ │ ├── tts_tools.py # Text-to-speech (12 tools)
│ │ ├── music_tools.py # Music generation (12 tools)
│ │ └── ... # Browser, vision, GGUF, codebase, etc.
│ └── routes/ # REST API endpoints
│
├── providers/ # LLM provider integrations
│ ├── llamacpp_provider.py # Local GGUF via llama-cpp-python
│ ├── lmstudio_provider.py # LM Studio HTTP API
│ ├── ollama_provider.py # Ollama HTTP API
│ ├── vllm_provider.py # vLLM OpenAI-compat API
│ └── openrouter_provider.py # OpenRouter cloud API
│
├── ui/ # Frontend (vanilla JS, no build step)
│ ├── index.html # Single-page application
│ ├── js/ # 18 ES modules
│ └── styles.css # All styles
│
├── modules/ # External engines (auto-cloned on first use)
│ ├── comfyui/ # → rookiemann/comfyui-portable-installer
│ ├── tts/ # → rookiemann/portable-tts-server
│ └── music/ # → rookiemann/portable-music-server
│
├── models/ # Downloaded model files
│ └── gguf/ # GGUF models (searched + downloaded from HuggingFace)
│
├── python/ # Portable Python 3.14.2 (auto-installed)
├── node/ # Portable Node.js 24.12.0 (auto-installed)
├── node_modules/ # n8n and npm packages (auto-installed)
│
├── manual/ # User manual (4,000+ lines, 89 screenshots)
│ ├── AgentNate-Manual.md # Full documentation source
│ ├── AgentNate-Manual.pdf # PDF version (140+ pages)
│ ├── build_pdf.py # MD → PDF converter
│ ├── build_wiki.py # MD → GitHub Wiki splitter
│ └── screenshots/ # All manual screenshots
│
├── data/ # Runtime data (conversations, settings, manifests)
├── settings/ # Configuration manager
├── workflows/ # Pre-built workflow templates
└── orchestrator/ # Request queue and health monitoring
AgentNate agents aren't simple chatbots. They have access to 187 tools across 19 categories and can autonomously chain them together to accomplish complex tasks.
| Category | Tools | Examples |
|---|---|---|
| Chat & Reasoning | 8 | Conversation management, system prompts, context control |
| Web & Research | 12 | DuckDuckGo search, URL fetching, content extraction, web scraping |
| Code Execution | 6 | Run Python/JS/Bash, manage virtual environments, install packages |
| File Management | 10 | Read, write, search, organize files on disk |
| Browser Automation | 8 | Navigate pages, click elements, fill forms, take screenshots |
| Vision & Analysis | 6 | Describe images, extract text, compare visuals, detect objects |
| n8n Workflows | 19 | Build, deploy, activate, execute, monitor, delete workflows |
| ComfyUI Creative | 36 | Generate images/video, build pipelines, manage models, batch process |
| Text-to-Speech | 12 | Generate speech, clone voices, stream audio, manage engines |
| Music Generation | 12 | Compose music, generate sound effects, manage models |
| GGUF Models | 5 | Search HuggingFace, download GGUF models, track progress |
| Sub-Agents | 8 | Spawn workers, delegate tasks, coordinate parallel work |
| Memory | 6 | Persistent memory across conversations |
| Codebase | 11 | Self-awareness via structured manifest |
| And more... | Marketplace, routing, settings, GPU monitoring |
You: "Generate a landscape painting of mountains at sunset,
then create a workflow that emails me when new images are generated"
Agent: [Thinking] This requires two tasks: image generation and workflow creation.
I'll delegate image generation to a sub-agent and build the workflow myself.
→ spawn_agent("ai_creative", "Generate mountain sunset landscape painting")
→ comfyui_generate_image(prompt="majestic mountain range at golden hour sunset...")
→ comfyui_await_result(instance_id=1, prompt_id="abc123")
→ build_workflow(nodes=[webhook, comfyui_trigger, email_send], deploy=true)
Agent: Done! I've generated your landscape painting (saved to ComfyUI output folder)
and deployed an n8n workflow that watches for new generations and sends an email.
AgentNate supports loading models from multiple providers simultaneously. Mix local and cloud, switch mid-conversation, or compare them in LLM Lab.
| Provider | Type | GPU Required | Notes |
|---|---|---|---|
| llama.cpp | Local GGUF | Yes (CUDA) | Direct model loading, full control over GPU layers and context |
| LM Studio | Local HTTP | Depends | Connect to running LM Studio instance, any model it supports |
| Ollama | Local HTTP | Depends | Connect to running Ollama instance |
| vLLM | Local HTTP | Yes | High-throughput serving, continuous batching |
| OpenRouter | Cloud API | No | Access to 200+ models (GPT-4, Claude, Gemini, Llama, etc.) |
Click to expand screenshots
 |
 |
| Chat with streaming responses | Agent executing tools autonomously |
 |
 |
| Side-by-side independent conversations | Compare models head-to-head |
 |
 |
| ComfyUI instance management | AI-generated artwork |
 |
 |
| Text-to-Speech with voice cloning | Music generation studio |
 |
 |
| n8n workflow management | Real-time GPU monitoring |
- User Manual (Wiki) -- Full documentation with 89 screenshots covering every feature
- PDF Manual -- 140+ page printable manual (included in repo)
- API Docs -- Interactive Swagger UI (available when server is running)
- OpenAI-Compatible API -- Drop-in replacement for OpenAI SDK clients
AgentNate achieves true portability by bundling self-contained runtimes that never touch the host system:
| Component | How | Why |
|---|---|---|
| Python 3.14.2 | Embeddable distribution with ._pth configured for site-packages |
No system Python, no PATH, no registry |
| Node.js 24.12.0 | Pre-built binary extracted to node/ |
No system Node, no nvm, no global installs |
| n8n | Installed via npm into local node_modules/ |
Workflow engine stays in the folder |
| Playwright | Chromium installed to python/.playwright-browsers/ |
Browser automation without system Chrome |
| llama-cpp-python | Pre-built CUDA wheel from custom build | GPU inference without building from source |
| ComfyUI | Managed in modules/comfyui/ with portable Git |
Full creative pipeline, isolated |
| TTS / Music | Managed in modules/tts/ and modules/music/ |
Audio engines with their own Python environments |
Everything lives inside the AgentNate folder. Copy it. Move it. Back it up. Put it on a different machine. It just works.
AgentNate's creative and audio capabilities are powered by three companion projects that are automatically cloned on first use. When you enable ComfyUI, TTS, or Music from the UI, AgentNate pulls the corresponding repo into modules/ and runs its installer — no manual setup required.
| Module | Repository | What It Provides | Installed To |
|---|---|---|---|
| ComfyUI | comfyui-portable-installer | Multi-GPU ComfyUI with 101 pre-built models, custom nodes, multi-instance management | modules/comfyui/ |
| TTS | portable-tts-server | XTTS v2, Fish Speech, Kokoro TTS engines with voice cloning and real-time streaming | modules/tts/ |
| Music | portable-music-server | Stable Audio Open and ACE-Step music generation with CLAP scoring | modules/music/ |
Each module is fully self-contained with its own portable Python environment, dependencies, and models. They run as separate processes managed by AgentNate's backend — start, stop, health check, and proxy are all handled automatically.
Already have these installed standalone? AgentNate will detect an existing installation in
modules/and use it directly. You can also point to a custom path in Settings.
AgentNate builds on and integrates these projects by the same author:
| Project | Description |
|---|---|
| auto-portable-python-deployer | The tool that generates portable Python deployment packages (used to bootstrap AgentNate) |
| comfyui-portable-installer | One-click portable ComfyUI with multi-GPU, 101 models, custom nodes |
| LocalSoundsAPI | Portable all-in-one audio studio (TTS, music, transcription, video production) |
| portable-tts-server | Standalone portable TTS server (XTTS, Fish Speech, Kokoro) |
| portable-music-server | Standalone portable music generation server (Stable Audio, ACE-Step) |
| llama-cpp-python-py314-cuda131-wheel | Pre-built CUDA wheel for llama-cpp-python on Python 3.14 |
| n8n-python-portable | Portable n8n with embedded Python runtime |
| vllm-windows-build | vLLM compiled for Windows with CUDA support |
AgentNate is built on the shoulders of incredible open-source projects:
Core Infrastructure
- FastAPI -- High-performance async Python web framework
- n8n -- Fair-code workflow automation engine
- ComfyUI -- Modular Stable Diffusion pipeline
- llama.cpp / llama-cpp-python -- Local LLM inference
LLM Providers
- LM Studio -- Local model management and serving
- Ollama -- Run LLMs locally
- vLLM -- High-throughput LLM serving
- OpenRouter -- Unified API for 200+ models
Creative Engines
- Stable Diffusion / Flux -- Image generation models
- Stable Video Diffusion -- Video synthesis
- XTTS v2 -- Multilingual voice cloning
- Fish Speech -- Fast expressive TTS
- Kokoro -- Lightweight English TTS
- Stable Audio Open -- Text-to-music generation
- ACE-Step -- Advanced music generation
Tools & Libraries
- Playwright -- Browser automation
- DuckDuckGo Search -- Web search (no API key required)
- Pydantic -- Data validation
- httpx -- Async HTTP client with streaming and resume support
- PyMuPDF -- PDF text extraction for RAG
- HuggingFace -- Model hub and GGUF repository
Built by one person with an AI assist, made possible by thousands of open-source contributors.
Contributions are welcome! See CONTRIBUTING.md for guidelines.
- Bug reports: Open an issue
- Feature requests: Open an issue
- Questions & ideas: Start a discussion
This is an actively developed project. New features ship regularly and feedback directly shapes the roadmap.
MIT -- Use it, modify it, ship it. Just include the license.