The nanoodle guide
How the pieces fit together — and why there is no server in any of them.
nanoodle is a node-graph playground for AI models that runs entirely in your
browser. You wire text, image, video, audio and LLM nodes into a workflow on a
canvas, hit Run, and each node calls NanoGPT
directly from your browser on your own API key. Any workflow can be turned into a
standalone app with auto-generated inputs, shared as a URL, or exported as a single
self-contained .html file.
There is no backend, no account system, and no analytics — the site is a handful of static HTML pages, and the source repository (MIT) is exactly what nanoodle.com serves. Your API key and your workflows live in your browser's localStorage; model requests go straight from your machine to nano-gpt.com. These pages document the parts you might want to build on: the workflow file format, the headless executors, the share-link mechanics, and how to host it yourself.
Pages in this guide
noodle-graph.json: nodes, links, ports, fields — every
key grounded in the editor's serializer.
Run workflows headlessly
Re-run a saved graph from Node.js or Python with the zero-dependency
nanoodle packages — library and CLI.
How share links work
The whole workflow or app rides in the URL fragment (#g= /
#a=) — which browsers never send to any server.
Self-hosting
It's a static site: clone the repo, serve the folder, done. The repo is the
privacy proof.
Elsewhere
- The editor — build and run workflows (also at
/app). - The app builder — turn a workflow into a shareable app.
- github.com/nanoodlecom/nanoodle — the site's source (MIT).
- nanoodle-js and
nanoodle-py — headless executors (npm / PyPI, both named
nanoodle). - /llms.txt — a machine-readable summary of the site for LLMs.
- Terms, Privacy & FAQ.