Agent Builders
Design multi-step LLM agents that reason, call tools, and stream progress.
- Planning + execution in one workflow
- Preview nodes to debug intermediate steps
- Trigger runs from Global Chat or CLI
Local-first AI workflow builder
Build AI workflows visually. Deploy anywhere.
NodeTool lets you compose text, audio, video, and automation nodes on a single canvas, run them on your machine, then ship the identical workflow to RunPod, Cloud Run, or your own infrastructure.
Start Here
NodeTool is the local-first canvas for building AI workflows—connect text, audio, video, and automation nodes visually, then run them locally or deploy the exact same graph to RunPod, Cloud Run, or your own servers.
Who uses NodeTool?
Knowledge & RAG Teams
Index private corpora, run hybrid search, and ground every answer in sources.
- Document ingestion + retrieval on one canvas
- Built-in ChromaDB collections
- Automations via Mini-Apps or APIs
Multimodal Makers
Prototype creative pipelines mixing audio, vision, video, and structured tools.
- Mix local diffusion with hosted APIs
- Scriptable data prep & charting nodes
- Deployable without rewriting code
What you can build right away
LLM Agents with Tool Access
Plan, call tools, and summarize results with streaming progress updates.
Agent pattern → Realtime Agent example →Retrieval-Augmented Generation
Ingest PDFs, chunk text, and answer questions grounded in citations.
RAG pattern → Chat with Docs example →Audio + Video Pipelines
Transcribe meetings, remove silence, add subtitles, or narrate generated imagery.
Transcribe Audio example → Story to Video example →Data Automation & Visualization
Fetch data, transform it with AI nodes, and publish dashboards or reports.
Data Viz pipeline → Data processing pattern →Your first 10 minutes
- Download NodeTool — install the desktop app for macOS, Windows, or Linux.
- Launch and pick models — install GPT-OSS + Flux in Model Manager for fast local runs.
- Open the Creative Story Ideas template — inspect nodes, press Run, and watch Preview stream results.
- Save & run from Global Chat — trigger the workflow directly from a chat thread.
- Publish as a Mini-App — hand teammates a form UI powered by the same workflow.
Choose your path
- I’m new to NodeTool: Getting Started, Workflow Editor, Tips & Tricks, Example gallery.
- I deploy NodeTool: Deployment Guide, Self-Hosted Deployment, Proxy Reference, Security Hardening, Terminal WebSocket, Storage.
- I build or integrate: Developer Guide, Workflow API, Chat Module + Chat API, CLI Reference, Architecture, Testing Process.
- Reference shortcuts: Cookbook Patterns, Glossary, Indexing, Asset Management, Node Packs.
Local-first or cloud-augmented
Local-only mode
All workflows, assets, and models execute on your machine for maximum privacy.
- Use MLX, llama.cpp, Whisper, and Flux locally
- Store assets on disk or Supabase buckets you control
- Disable outbound traffic entirely if needed
Cloud-augmented mode
Mix local nodes with OpenAI, Anthropic, or RunPod workers when you need extra capacity.
- Configure API keys in Settings → Providers
- Deploy the same workflow to RunPod or Cloud Run
- Automate runs through the Workflow API or chat APIs
Why teams choose NodeTool
Privacy-first
Run LLMs, Whisper, and diffusion models locally without shipping data to third parties. Opt into APIs only when needed.
Single workflow, many surfaces
Create once in the editor, trigger from Global Chat, expose via Mini-Apps, or call it from the Workflow API—all backed by the same graph.
Deploy without rewrites
When you outgrow your laptop, push the same workflow to RunPod or Cloud Run. No refactoring required.
Popular destinations
Build with the community
NodeTool is open-source under AGPL-3.0. Join the Discord, explore the GitHub repo, and share workflows with other builders.