NodeTool: Open-Source Visual AI Workflow Builder Platform, Run 500K+ Models Locally

NodeTool is an open-source visual AI workflow building platform that enables connecting LLMs, generating media, and building agents through drag-and-drop nodes, with no coding needed to run locally or in the cloud. The platform integrates Ollama and MLX to support over 500,000 models, is compatible with mainstream APIs like OpenAI and Anthropic, and provides cross-platform support for desktop, Web, CLI, and mobile. Its core advantages include local-first approach, multimodal processing, and AI agent system, etc.

NodeTool's core philosophy is "AI should run on your machine, right next to your data", responding to needs for data privacy, cost control, and flexible deployment. The project uses the AGPL-3.0 open-source license, is maintained by developers like Matti and David, has an active Discord community, and adopts a monorepo architecture, including TypeScript backend, React frontend, Electron desktop shell, and React Native mobile app.

• Visual Workflow Builder: Drag-and-drop nodes to create type-safe connections without coding
• Local-First AI Support: Natively supports Ollama, MLX (Apple Silicon), and GGUF/GGML formats, protecting privacy and reducing costs
• Massive Model Ecosystem: Integrates over 500,000 models from HuggingFace, covering multimodal tasks and compatible with mainstream cloud APIs
• AI Agent System: Built-in 100+ tools, supporting task planning, tool calling, and multi-step reasoning
• Multimodal Processing: Unified architecture for processing text, images, video, and audio
• Real-Time Streaming Output: Asynchronous execution with real-time preview of intermediate results
• Flexible Deployment: Supports multiple methods like Docker and RunPod
• Code Extension: Custom nodes in Python/TypeScript

NodeTool adopts a modular monorepo architecture:

• Backend Core: Includes 28 packages such as kernel (DAG engine), node-sdk (node registration), base-nodes (100+ built-in nodes), agents (agent system), runtime (LLM integration), etc.
• Frontend: Built with React+Vite+MUI+React Flow for an intuitive editing interface
• Desktop App: Electron-packaged cross-platform program
• Mobile App: Extended to mobile via React Native+Expo

• Video Generation: OpenAI Sora 2 Pro, Google Veo 3.1, xAI Grok Imagine, etc.
• Image Generation: Black Forest Labs FLUX.2, Google Nano Banana Pro, DALL-E 3
• Audio Processing: OpenAI Whisper (speech recognition), OpenAI TTS, ElevenLabs
• Text Models: GPT-4, Claude, Gemini, Llama, Mistral (local/cloud) Users can select models via corresponding nodes; some require API keys, while others are accessible through the kie.ai aggregation service.

• LLM Agent Applications: Customer service, data analysis, content creation
• Creative Content Pipeline: Automatic conversion from copy to video
• RAG Knowledge System: Enterprise knowledge base, customer service robot
• Data Conversion Workflow: Large-scale data cleaning, ETL automation
• Mini Apps: Package workflows as Web apps for sharing
• Automation Pipeline: Content moderation, intelligent archiving

NodeTool fills the gap between ComfyUI (focused on media generation, limited local LLM/agent support) and n8n (focused on business automation, basic AI agents without local LLM support). Its advantage lies in the comprehensive combination of general AI workflows, local model operation, agent system, and multimodal processing.

System Requirements:

• Windows: NVIDIA GPU (4GB+ VRAM), 20GB disk space
• macOS: M1+ chip, 16GB+ RAM
• Linux: NVIDIA GPU (4GB+ VRAM), Flatpak build Quick Start: Set up environment with conda → Install core packages → Run backend (port 7777) and frontend (port 3000). Documentation is available at docs.nodetool.ai

Conclusion: NodeTool represents the direction of AI workflow tools. By leveraging visualization and local-first approach, it lowers the development threshold, allowing more users to use AI to solve practical problems.