Your models, your agents, your files.
AI comes in three ways: an assistant that builds with you, generation as nodes in the graph, and a bridge for the agent you already use. Then the artwork you already have comes in as layers.
It builds the graph with you — and can't invent nodes.
The assistant works on the live graph while you watch, and a whole turn is a single undo — so trying something it suggests costs one keystroke to reverse.
It can't invent nodes
It builds against a closed registry of the node types the engine actually has. It cannot hallucinate an operator that doesn't exist, because there is nowhere for one to come from.
It knows what you're looking at
The pane you're in, what's selected, and the one-hop neighbours with their slot names — so "wire this into that" means something.
On your own key
Paste an Anthropic key and nooodlebot talks to the model straight from your browser. The key stays in the session and never enters the saved file.
Generation is an operator with typed inputs.
A generation node is fed by your own work — a model you built, a layout you composed, a mask you painted — and its output is wired into whatever comes next. That one architectural choice is the difference between art direction and a slot machine.
One node per backend
Providers are transport; models are swappable. The model list is read live from the endpoint rather than hard-coded, so a new model appears without waiting for us.
Your key, or your hardware
Paste an Anthropic key and generation runs from your browser with no backend of ours in the path — the key lives in session storage and never enters the saved file. Or point it at Ollama, LM Studio or llama.cpp and the model runs next to the material instead of the material travelling to the model.
The backends
Ollama, LM Studio, llama.cpp, Anthropic, OpenAI, OpenRouter and any OpenAI-compatible endpoint for text. Gemini and a local diffusers server for images and video, plus ComfyUI for your own workflows.
Realtime diffusion
StreamDiffusion over a persistent socket, running against a live feed rather than one still at a time.
Let your own agent drive the editor.
Point Claude Code at the tab in front of you: 50 tools to create and wire nodes, set keys, drive the transport, read cooked geometry and read the running state of a machine. It can ask for the current preview and iterate against what it sees.
Runs on your machine
The bridge is a small local server that sits next to your MCP client. The editor tab connects to it only once you switch it on.
The same tools as nooodlebot
An outside agent gets the tool set the assistant uses, plus a way to pick which open tab it's working in.
We orchestrate models. We don't train them.
Training foundation models is a capital race that obsoletes itself every few months. We'd rather be the best place to use whatever exists.
Model-agnostic by construction
Providers are transport. Models are swappable without re-architecting your graph, so when a better one ships you point at it — and when one degrades or changes terms, you aren't stranded.
Open weights, so the cost curve works for you
Open models get cheaper and better without us doing anything. We ride that curve rather than paying to push it, which is also why we can afford a free tier that isn't crippled.
Runs where you choose
In the cloud on a key you paste, so nobody needs a GPU to start, and on your own hardware when the work can't leave the building — the case that matters for client IP under NDA. See motion design studios for how that's set up.
No lock-in on the way out
A scene compiles to an ordinary web page and a workflow hands you ordinary files. The graph is yours, and so is what it makes.
Start from the files you already have.
Nobody starts here. The work arrives as a layered artwork file, a model somebody else built, or a folder of images — so those come in as operators you can keep working on, rather than as a flat picture of themselves.
Layered artwork stays layered
A PSD or an Illustrator file comes in as an editable layer graph, not a flattened image. The layers are nodes, so the thing your designer sent is the thing you keep art-directing.
Figma Coming soon
Bring a Figma design in the same way: as layers you keep working on, not a picture of them.
Vector as real geometry
SVG arrives as paths you can extrude, bevel and deform — a logo becomes 3D geometry rather than a texture of a logo.
Models you can still deform
GLB and glTF come in as geometry operators, with the deformer stack available on top. A GLB's baked materials can be exploded into editable material nodes.
Lighting from a plate
HDRI environments for image-based lighting, with the environment rotatable — so a product sits in a room rather than under a default key light.
Your fonts, not a web list
Text reads the fonts installed on your machine, so a brand face is available without a workaround.
What is not here yet
There is no importer yet for another tool's timeline or node graph — you rebuild the system once, and the versions after that are free. Text in a PSD or Illustrator file arrives as pixels for now, not as live type.
Wire it up and see the difference.
Free to start, nothing to install. The gap between a panel and a node is obvious about ninety seconds after you've used one.
The model you want wired in, and the one you can't use
Provider coverage, local weights, what your client contracts permit — tell us and we'll wire a step with you. This is the fastest-moving part of the engine, so it's where your input lands soonest.
No spam. A real reply from a human.