NodeTool

What it is

NodeTool is an open-source visual canvas for building AI workflows that mix text, image, video and audio. The company describes it as "an open-source canvas for image, video, audio, and text that runs on your own keys", and its comparison pages pitch it as the chatbot and agent builder alternative you can host yourself. The homepage says it is free, licensed under AGPL-3.0, and available for macOS, Windows and Linux, and that it is built alongside working artists, designers and video editors.

The reader it suits best is a maker, a creative team or a developer who wants control of models, keys and files. That includes someone who wants a chatbot or document-search agent, but the pages are candid that the product's center of gravity is media creation. The pages held here do not say where the company is based. The homepage lists a GitHub link and a team contact section, but gives no location.

What you build and how

The workspace is agent-first. The homepage says you give an agent a brief, and it "drafts the script, boards the shots, renders the takes, and cuts them on a timeline". What comes back is a project you can open and change yourself, or hand back to the agent. The workspace itself covers script, storyboard, sketch, timeline and 3D, with built-in editing tools such as masks, inpaint, relight, upscale and layers.

Underneath is a node graph. When a project comes out right, the company says you keep the process that made it: chain models, editors and code into a graph you can re-run on new inputs, ship as an app, or drive headless from the CLI. The node catalog is described as hundreds of blocks for models, data and files, plus one for every model on Replicate, fal.ai and Kie. Templates and recipes ship as workflows you can open, run and change, and a recipe chains several into one job. The homepage's example projects include building a catalog set from one product photo, revoicing an English presenter in Spanish, and a 6-shot storyboard with a shared cast.

2 pieces make the canvas usable by other people. The app builder gives a workflow a screen with inputs, a Run button and the result, so a teammate who never sees the canvas can run it. And the developer tools, an SDK, a CLI and an MCP server that Claude Code and Cursor can call, bring your own code onto the canvas.

For the chatbot reader, the relevant claim is on the Dify and Flowise comparison pages. The company says NodeTool covers the same agent and document-search ground as those tools, on a visual canvas, with agent and knowledge-base nodes on the same canvas as image, video and music generation. It also says an agent can wire the graph for a chatbot and its media pipeline, validate it and repair what fails. The pages do not describe a chat widget to embed on a website, channel connectors such as WhatsApp or Slack, or a hosted chat endpoint, so a reader who needs a deployed support bot should check how a NodeTool workflow would be exposed. The app builder screen is the described way to hand a workflow to users.

For the open-source low-code reader, the points are concrete. The homepage says "free and open source, AGPL-3.0", and a comparison page calls AGPL-3.0 an OSI-approved license. Scripts, storyboards, sketches, timelines and the multi-track cut are "ordinary files on your disk", and you can export a .nodetool bundle and open it anywhere. The comparison table on the homepage says the product runs as a desktop app and in the browser, with "self-host any time", against hosted tools that run on "their servers only".

Model access is on your own accounts. The company says you route shots through video, image, audio and language models from every major provider on your own keys, and open weights run on your machine through MLX, Ollama and llama.cpp. Its example is a video model that FAL, Replicate and KIE each sell at a different price, where NodeTool lets you pick the cheapest. The desktop app keeps files, models and keys on disk and runs offline, and the homepage says local projects and supported local models can remain available without a network connection.

What the pages say about cost and editions

There is no subscription tier on the pages held here. The company says you pay providers directly at their list prices, with "no credits, no markup", on self-hosted and on NodeTool Cloud use alike. The homepage points to a pricing page for what a month costs "model by model", and that page was not available here, so no dollar figures appear in this entry. The practical budget is your provider bills, which grow with every render, not a flat license.

2 editions are described. Studio is the desktop app, which runs on your computer. Cloud is a browser version that needs nothing installed, and the company labels it alpha and says to use it for evaluation and lightweight access. Treat Cloud as a preview, not a production host.

The comparison pages also describe 2 other products, and these are NodeTool's statements about competitors. They say Dify is published under a modified Apache 2.0 license with commercial limits and sells cloud plans on seat and usage terms, and Flowise source-available under Apache 2.0 with a hosted cloud sold on usage-based credits. Check each project's own license before relying on those summaries.

Who it suits and who it does not

NodeTool suits a creative team or a freelancer producing video, image and audio work who wants model flexibility and files they own. It suits a developer who wants a self-hostable, open-source canvas with an SDK and CLI. And it suits a small team that wants an agent over its documents next to media generation on one canvas, with the app builder as the way to hand a finished workflow to non-technical colleagues.

It fits less well if your only goal is a text support bot on a website. The company itself says Dify's prompt management, observability and app-store-style deployment make it the pick for a text-first language model app, and that Flowise is the quickest route to a LangChain chatbot over a document set. It also asks you to manage your own provider accounts and billing, and the browser edition is in alpha. A node canvas is more machinery than a text-only workflow needs.