The open creative AI workspace

Ask for it. Then take the controls.

Describe what you want and NodeTool's agent builds it: a workflow, a layered image, a video cut, a small app. Everything it makes lands on a canvas you can open, inspect, and change by hand. Your keys or your own hardware. Open source, AGPL-3.0.

NodeTool canvas
Get started Examples Cookbook

Start by asking

One agent runs through the whole app. It reads the document you have open and edits it with the same actions you would use by hand, so its work is a normal workflow, sketch, or timeline afterwards. Nothing it builds is locked, and nothing it builds is hidden.

Say what you want

"Turn this logline into a storyboard and a cut teaser." The agent picks nodes, wires them, chooses models, and runs it.

Chat & Agents →
Watch it work

Every step streams: the plan, each tool call, each node as it lights up. A permission chip sets whether it asks first.

The agent loop →
Take over any time

Click a node, change a value, rewire an edge, re-run. The graph is the source of truth for both of you.

Workflow editor →

The same agent works on every surface: the node graph, the Sketch Editor, the Video Editor, scripts and storyboards, and the App Builder. Ask it to add a track, paint a layer, revise a shot, or wire a form field, and watch the document change.

Three flagship workflows, end to end. Each starts from a few inputs and builds a finished result on one canvas you can re-run, restyle, and re-point at your own story. See all use cases →

What you can do

  • Have the agent build it — Describe the outcome and it assembles the graph, picks the models, runs it, and reports what came back.
  • Edit what it built — Click a node, change a value, re-run. Nothing is generated behind glass; the graph is a file you own.
  • Put agents inside the pipeline too — An Agent node plans, calls tools, and streams, as one step of a larger workflow.
  • Mix models from every vendor — Wire Flux next to GPT-5.6 next to ElevenLabs in one graph. Pick the best model per step, not per project.
  • Run frontier models locally — Ollama, MLX, and GGUF on your hardware. Works offline. Files never leave your disk.
  • Bring your own keys — Pay OpenAI, Anthropic, Gemini, Replicate, FAL, and ElevenLabs directly. No credit markup, no provider tax.
  • Ship a workflow as a Mini-App — Hide the graph, expose just inputs and outputs. Share a link, no install required.
  • Chat with your documents — Local SQLite-vec, embeddings, RAG. Your data never leaves the machine.

Studio or Cloud

Same code, same workflows. Both AGPL-3.0.

NodeTool Studio — desktop

Mac, Windows, Linux. Local inference via Ollama, MLX, and GGUF. Works offline. Prompts and outputs stay on disk. BYOK for cloud providers when you want them.

Download Studio →
NodeTool Cloud — browser

Hosted, no install. Same canvas, same nodes. BYOK for every cloud provider — OpenAI, Anthropic, Gemini, Replicate, FAL, ElevenLabs, HuggingFace. No local models.

Open Cloud →

No credit markup. Cloud hosts the same code in this repo. Self-host the Docker images any time. You pay providers directly.

What you can build

Image and video

Flux, Qwen, Wan, Seedance, Sora, Veo, Kling on one canvas.

Movie Posters →
Agents

Agents that plan, call tools, and drive pipelines — in chat or as a node.

Fetch Papers →

More patterns — pipelines, data, RAG, email — in the Cookbook.

Get started

  1. Download NodeTool for macOS, Windows, or Linux.
  2. Connect a provider and ask the agent for something.
  3. Open what it built, change it, ship it as a Mini-App.

Explore

Open source

AGPL-3.0. Discord · GitHub.