For people choosing an agent-driven creative workspace with nodes underneath. NodeTool runs on the machine you install it on; Kaitoi hosts the workspace and runs local models and code on whichever machine you connect.
NodeTool is an open-source creative workspace, under AGPL-3.0, with an agent that works inside real editors: node workflows, storyboards, scripts, a timeline, sketches, 3D scenes and games. Its main edition is a free desktop app that runs local models or your own provider keys.
Kaitoi covers much of the same ground: an assistant, Penny, that builds and edits node graphs, storyboards and timelines, with 3D tools, a drawing canvas and local models. The difference is where things run. Kaitoi Studio is hosted, with more than 1,400 generative models available on credits as of September 2026, and Kaitoi CLI connects your own computer or a GPU machine elsewhere, to run local models and your own code.
Checked against each product's own documentation on .
NodeTool is free and open. The desktop app runs on macOS, Windows and Linux, the source is AGPL-3.0, and it can run as a Docker container on a machine or server you control.
It is local-first. It runs models through Ollama, MLX, llama.cpp, vLLM and LM Studio, connects to hosted providers on your own keys with no markup, mixes local and hosted models node by node, and can work offline with local models.
Its agent is thorough. It edits the same documents you open, works in Plan, Default or Auto permission modes, can be held to a spend cap per run, and logs every tool call so you can stop it and take over.
It has a full short-film pipeline. A Director agent turns a brief into a screenplay, stills and then clips, a script editor casts a voice per character, the timeline carries video, audio, overlay, subtitle and MIDI tracks, and a cut can be reframed to 9:16, 1:1 or 4:5. It can also run exported ComfyUI workflows as a node.
Free to start, with open source coming. Kaitoi is free to start with no subscription, and hosted models cost the provider's own price with no markup. Studio is hosted in the browser, and Kaitoi Open Source, a distribution you run yourself, is announced as coming soon.
Local where you want it. Kaitoi CLI connects Ollama, ComfyUI, Blender and Runtime recipes on your own computer or a GPU machine elsewhere to Studio, mixed with hosted models in one graph, on credits or your own keys. Kaitoi Code runs on Ollama, LM Studio or any OpenAI-compatible server with no Kaitoi login; Studio itself needs a connection.
An agent that shows its work. Penny builds and edits graphs, writes node code, lays out storyboard frames, places and splits timeline clips and arranges Lab boards; each reply lists the tools it used, and a stop icon ends the request. Penny gives a cost estimate when asked but has no permission modes or enforced cap per request; Kaitoi Code, in the terminal, has Plan, Build and Crew modes and a credit cap.
A brief-to-cut pipeline. AdMuse concepts and scripts feed storyboard shots, clips from Kling, Veo or Seedance through Lab Skills, and the timeline, which can output 9:16, 1:1 or a custom size; Gemini voice nodes voice dialogue between speakers; and on Studio an exported ComfyUI workflow runs as one node. Subtitles are rendered by a node and placed as a video clip; there is no script editor, no text, overlay, subtitle or MIDI track, and no one-step reframe of a whole cut.
Beyond that, Studio has a drawing canvas for sketches, masks and crops, though Penny does not draw on it. A node can carry its own HTML, CSS and JavaScript interface, WebAssembly included, so a node can be anything up to a playable game, which a Station can ship to visitors as its app.
Where Kaitoi differs is the split between the workspace and the hardware. In NodeTool the whole app runs where you install it, on a desktop or a server. In Kaitoi the workspace is hosted, so there is nothing to operate, and Kaitoi CLI runs wherever the models should run: the same laptop, a workstation at the studio or a rented GPU. You can drive Studio from a laptop while a bigger machine does the work.
On that machine, open models and your own code can become nodes. Kaitoi Code, the coding agent in Kaitoi CLI, takes a Hugging Face model or a GitHub repository, writes a Runtime recipe, tests it and, on the Studio plan, creates the node.
In the cloud, Kaitoi's providers offered more than 1,400 generative models and around 400 language models as of September 2026, on credits with no keys to set up or on your own keys. Many already have nodes, and the New Models panel has Penny turn a model that has no node yet into one, which uses credits for Penny's work. Local and hosted models sit in the same graph.
Outside agents work in Kaitoi through its MCP server on every plan, and on Studio they can write nodes. When a step is missing, Penny writes it as a node on every plan, and nodes and templates can be published to Community. Finished work goes out as a Station, a simple form or a custom app where each visitor pays for their own runs, or as an API endpoint.
| Question | NodeTool | Kaitoi |
|---|---|---|
| Where it runs What you install, and who operates it. | A free desktop app for macOS, Windows and Linux, or a Docker container on your own machine or server. A browser edition, NodeTool Cloud, is in alpha. | Studio is hosted in the browser. Kaitoi CLI runs on your own computer or a GPU machine elsewhere and connects it to Studio, so the models run there while you work from a laptop. Kaitoi Open Source, a distribution you run yourself, is announced as coming soon. |
| Local models Running models on your own hardware. | Ollama, MLX, llama.cpp, vLLM and LM Studio, mixed with hosted models node by node, and offline work with local models. | Ollama, ComfyUI and Blender through Kaitoi CLI, and Hugging Face models or GitHub repositories as Runtime recipes that Kaitoi Code writes and tests, mixed with hosted models in one graph. Kaitoi Code runs on Ollama, LM Studio or any OpenAI-compatible server without a Kaitoi login; Studio itself needs a connection. |
| Hosted models How cloud models are reached and paid for. | Your own provider keys at the provider's price with no markup. In the desktop app the agent can also run on a ChatGPT or Claude subscription. | More than 1,400 generative and around 400 language models on Kaitoi credits with no keys to set up, at the provider's own price with no markup, or your own provider keys, billed by your provider, on every plan. Kaitoi Code can hand subagent roles to Claude Code or Codex under their own login. |
| Agent controls Permissions, spending limits and seeing what the agent did. | Plan, Default or Auto permission modes, a spend cap per run, and a log of every tool call so you can stop the agent and take over. | Penny lists every tool it used in each reply, gives a cost estimate when asked and can be stopped, but has no permission modes or enforced cap per request. Kaitoi Code has Plan, Build and Crew modes, optional auto-approval by a reviewer model, and a credit cap. Team members have weekly caps. |
| Assistant-made nodes The assistant adding a new reusable step. | The assistant writes sandboxed JavaScript scripts that can be added to the node menu. We found no documentation of it writing TypeScript or Python node packs. | Penny writes Python nodes, with their own HTML, CSS and JavaScript interface when wanted, saves them and adds them to the graph, on every plan. On Studio, outside agents can write nodes through MCP. Nodes can be published to Community. |
| Film production From a brief to a finished cut. | A Director agent from brief to screenplay, stills and clips; a script editor with a voice per character; video, audio, overlay, subtitle and MIDI tracks; and reframing to 9:16, 1:1 or 4:5. | Penny chains AdMuse concepts and scripts, storyboard shots, video Skills and the timeline, and Gemini voice nodes voice dialogue between speakers. Timeline clips are video, audio or images, and subtitles are rendered by a node and placed as a video clip; there is no script editor or text, overlay, subtitle or MIDI track. Output can be 9:16, 1:1 or a custom size, and a reframe node changes a clip's aspect with AI; there is no one-step reframe of a whole cut. |
| Handing off to an editor Moving an edit into Premiere, Resolve or similar. | Exports MP4, WebM, ProRes MOV and PNG sequences. Premiere handoff through Final Cut Pro 7 XML is partial, and Final Cut Pro and DaVinci Resolve handoff are not yet available. | Renders MP4, WebM, or MOV in ProRes 422 HQ (4444 XQ for HDR), with no image-sequence export, and exports and imports OpenTimelineIO with copies of the clip media for editors that read it. |
| 3D and splats Working with 3D models and Gaussian splats. | A 3D editor for glTF scenes, 3D generation through providers, and Blender rendering on local installs. We found no Gaussian splat support documented. | 3D generation through Tripo, Meshy, Hunyuan3D and World Labs, a Gaussian splat viewer, a 3D Editor node with render passes and Scene GLB export, and Blender rendering through a connected Blender. |
| Outside agents Working with the product from other AI tools. | NodeTool runs as an MCP server, so Claude Code, Codex, Cursor and other MCP clients can generate media and edit NodeTool documents, and its desktop agent can use tools from external MCP servers. | Kaitoi's MCP server, on every plan, lets compatible AI clients build graphs, run approved work and inspect results, and on Studio write nodes. Penny calls outside MCP servers, and a REST API runs graphs from your own code. |
| Sharing and teams Letting others run or see your work, and who pays. | Mini Apps can get a public link on a server in production mode, with runs on the owner's account within a spend budget. Servers with user accounts add view and edit share links. We found no team plan published. | Stations, a simple form or a custom app, where each visitor pays for their own runs and a Station that runs your own code is invite only; workflow API endpoints that run on your credits, with no budget per endpoint; read-only share links; and team workspaces with roles, shared projects and a pooled balance on a team Studio subscription. |
Each cell comes from the product's own website, documentation or code repository. "Not publicly documented" means we found no official page that says so, not that the feature is missing. Plans and features change; check the official pages before you decide.
A single image becomes an interactive space: Kaitoi separates environment and objects, rebuilds the scene as a navigable Gaussian splat, converts assets to 3D, adds physics and audio, and uses a viewer node generated by Penny for the workflow.
This film, from our Kaitoi Signals series, builds an explorable scene from a single picture, with a viewer Penny wrote for the job.
Read more in the documentation:
NodeTool Studio is free and open source, and NodeTool Cloud is free during its alpha, with future pricing not announced. Model costs are billed by your providers on your own keys.
Free to start, with no subscription. Credits pay for hosted models, including Penny's, at the provider's own price with no markup, and for paid Community nodes; steps that call neither, and models on your own machines, cost nothing. Your own provider keys work on every plan and are billed by your provider. Studio adds included credits, team seats and hand-edited code.
Yes. NodeTool Studio is free and open source under AGPL-3.0, and you pay your model providers directly. Kaitoi is free to start too, with no subscription: you spend credits only on hosted models, at the provider's own price with no markup, and on paid Community nodes, or use your own provider keys.
Yes. NodeTool runs exported ComfyUI workflows as a node. Kaitoi wraps an exported ComfyUI workflow as a node with the inputs and outputs you choose, on the Studio plan.
Yes. Kaitoi CLI runs on the machine with the GPU, your own computer or one elsewhere, and connects it to Studio, so Ollama, ComfyUI, Blender and Runtime recipes run there while you work in the browser. Kaitoi Open Source, a distribution you run yourself, is announced as coming soon.
Studio does not; it is a hosted service. Kaitoi Code can work on a local model with no Kaitoi login, and Runtime runs recipes on your machine, although installing a recipe downloads its dependencies and weights. NodeTool runs offline with local models.