For studios and artists comparing node-based AI workflow tools. Griptape fits a Nuke pipeline around an engine on your own machine; Kaitoi hosts the graph, connects your own machines, and hands finished work to editors, clients and other software.
Griptape is a node-based platform for orchestrating AI models and agents, aimed at studios and creative professionals, with a focus on VFX and animation pipelines. Foundry, the maker of Nuke, completed its acquisition of Griptape on 18 February 2026 and markets it as Griptape Nodes, the Griptape Framework and the Griptape Engine. The engine that runs workflows is open source under Apache 2.0, while the Griptape Nodes application around it is under Foundry's licence.
Kaitoi is a hosted node graph where more than 1,400 generative models, your own Python steps and your own machines work together, with a storyboard and a timeline that hands the cut to an editor through OpenTimelineIO. Penny builds the graph with you and writes the nodes that are missing. Kaitoi is free to start, with no subscription, and hosted models cost the provider's own price with no markup.
Checked against each product's own documentation on .
Made for VFX pipelines. A first-party Nuke library runs Nuke scripts from a workflow and publishes a workflow into Nuke as a versioned gizmo, using a local Nuke licence. First-party libraries also handle OpenEXR files and frame sequences, OpenColorIO colour management and Autodesk Flow Production Tracking, and Blender connects through a node library and an MCP server.
A node editor with a local engine. Griptape Nodes is a drag-and-drop node editor, as a desktop app for Windows, macOS and Linux or in a browser, that drives an engine on your own machine, with workflows and files kept in a local workspace by default. Open image and video models run on your own GPU, and hosted models from Google, OpenAI, Black Forest Labs, ByteDance, Runway, Luma, Kling, ElevenLabs and others run through Griptape's proxy on credits or on your own API keys. Custom nodes are Python, with JavaScript widgets, in libraries that install from Git into their own environments.
An agent and an MCP server. A built-in agent adds and connects nodes, runs the workflow and inspects results, asks before it runs workflows, generates images or uses outside tools, and follows the skill files a studio gives it. Griptape Nodes also runs an MCP server, so tools such as Claude Code or Cursor can build and run workflows, and the agent can connect to outside MCP servers.
Open source underneath. The engine that runs Griptape Nodes workflows is open source under Apache 2.0, and so is the Griptape Framework, a Python library for building AI agents and pipelines in code. Workflows are saved as plain Python files that can be kept in Git, and a Python scripting interface creates and runs them.
Publishing workflows. A workflow can be published as a folder that runs headless, or as a Griptape Cloud Structure that runs through an HTTP API, and pipeline scripts can trigger workflows on the engine over local HTTP.
Enterprise from Foundry. Griptape Enterprise, announced in July 2026, is custom-priced, with unlimited users, on-premises deployment through an Admin Server that is the single connection out of the studio network, seat management and Cedar permission policies.
Pipelines, through Blender, OpenEXR and OpenTimelineIO. Kaitoi CLI connects your own Blender, ComfyUI and Ollama to Studio, on your own computer or a GPU machine elsewhere, and Blender rendering runs through that connected Blender. Nodes read and write OpenEXR images, adjust their exposure and turn them into HDR video, and Color Space Convert moves between colour spaces such as ACEScg, Rec.709 and Rec.2020, without OpenColorIO configs. The timeline renders MP4, WebM or MOV in ProRes 422 HQ, or 4444 XQ for HDR, and exports and imports OpenTimelineIO for DaVinci Resolve or Premiere, and colour nodes handle levels, colour wheels, HDR grading, LUTs and white balance. Production trackers can be reached through your own code, such as a node calling their API, or through the REST API. Runway's Ruby node turns an SDR clip into an OpenEXR frame sequence, ACEScg included. Kaitoi has no Nuke plugin or connection and no built-in Flow Production Tracking connection, and the timeline itself renders no image sequence.
A node graph, hosted, with your own machines joining. Kaitoi Studio is a hosted node graph in the browser, and Kaitoi CLI brings in your own computer or a GPU machine elsewhere, so open models run there through your ComfyUI or as Runtime recipes, and on Studio an exported ComfyUI workflow becomes one node. More than 1,400 hosted generative models, among them OpenAI and Google Gemini models, FLUX, ByteDance's Seedream and Seedance, LTX video models, Runway, Luma's Uni-1, Photon and Ray models, Kling and ElevenLabs, run on credits at the provider's own price with no markup or, for providers that accept them, on your own keys. Nodes are Python and can carry their own HTML, CSS and JavaScript interface, WebAssembly included; Penny writes them on every plan, editing code by hand needs Studio or Enterprise, and on Studio outside agents write nodes through Kaitoi's MCP server. Hosted nodes use packages from an approved list, code that needs any other package runs on your machine as a Runtime recipe, and nodes and templates publish to Community. There is no desktop app, and projects live in the hosted Studio rather than a local workspace.
An agent and an MCP server. Penny builds and edits graphs, writes new nodes, edits storyboards, timelines and Lab boards, and follows the Skills and Playbooks you give it, on every plan; each reply lists the tools it used, and a stop icon ends the request. Kaitoi's MCP server, on every plan, lets compatible AI clients build graphs, run approved work and inspect results, and on Studio write nodes, and Penny calls outside MCP servers too. Studio shows a credit estimate before a node or graph runs, and Kaitoi Code, in the terminal, has Plan, Build and Crew modes.
Open source, coming soon. Kaitoi Studio is hosted, and Kaitoi Open Source, a distribution you run yourself, is announced as coming soon. Today Kaitoi Code, the coding agent in Kaitoi CLI, turns a Hugging Face model or a GitHub repository into a Runtime recipe, tests it and, on Studio, creates the node, and it can use Kaitoi workflows from a code project. A REST API runs graphs from your own code, with no published SDK, and the project keeps the graph, with templates and backups to return to an earlier version.
Publishing workflows, to code and to people. A saved template becomes an asynchronous API endpoint, whose results your code polls or streams and which runs on your own credits, and a REST API runs graphs from your own code. For people, a workflow becomes a Station at a stable link: a simple form or a custom app built from a node's own interface, where each visitor runs on their own Kaitoi account and pays their own costs, listed if you like in a reviewed gallery at kaitoi.io/stations. A Station that runs your own code, Penny-written nodes included, is invite only, and steps that need a local service such as your ComfyUI stay in Studio.
Teams and enterprise. On a team Studio subscription, team workspaces have roles that control what each member can do, shared projects, pooled credits and weekly caps per member, and every member takes a seat. Share links, on every plan, are read-only with optional cloning, and Kaitoi never trains models on your work. Studio is hosted on infrastructure Kaitoi operates, and a version you run yourself, Kaitoi Open Source, is announced as coming soon.
Beyond that, the project goes past the graph. A storyboard of scenes and shots feeds a multi-track timeline, Blocks shows the same nodes as a simpler chain, Lab keeps a canvas of references beside a conversation with Penny, and 3D work has a Gaussian splat viewer and a 3D Editor node with render passes.
| Question | Griptape | Kaitoi |
|---|---|---|
| Where the work runs What you install, and whose hardware runs the models. | A desktop app for Windows, macOS and Linux, or a browser editor, driving an engine on your own machine. A library's nodes can run on a more powerful computer, and workflows can burst to AWS Deadline Cloud. Workflows and files stay in a local workspace by default, and Enterprise can deploy on premises. | Studio is hosted and runs in the browser, with hosted models on credits. Kaitoi CLI connects your own computer or a GPU machine elsewhere, so ComfyUI, Ollama, Blender and Runtime recipes run there while you work from a laptop, and steps on your own hardware use no credits. There is no desktop app, and a version you run yourself, Kaitoi Open Source, is announced as coming soon. |
| Pipeline tools Working with the compositing, 3D and tracking tools a studio already runs. | A Nuke library runs Nuke scripts and publishes workflows as Nuke gizmos, with a local Nuke licence. Libraries read and save OpenEXR files and frame sequences, manage colour with OpenColorIO and connect to Autodesk Flow Production Tracking, and Blender connects through a node library and an MCP server. | Kaitoi CLI connects your own Blender for rendering. EXR nodes read and write OpenEXR images, adjust their exposure and turn them into HDR video, and Color Space Convert handles ACEScg, Rec.709 and Rec.2020, without OpenColorIO configs. The timeline exports and imports OpenTimelineIO for DaVinci Resolve or Premiere and renders ProRes 422 HQ, or 4444 XQ for HDR, and production trackers can be reached through your own code or the REST API. Runway's Ruby node turns an SDR clip into an OpenEXR frame sequence, ACEScg included. Kaitoi has no Nuke plugin or connection and no built-in Flow Production Tracking connection, and the timeline itself renders no image sequence. |
| Custom nodes Adding a step the product does not ship. | Python nodes with optional JavaScript widgets, in libraries that install from Git into their own environments, and a public directory of first-party and community libraries. Griptape documents writing nodes with coding assistants such as Cursor or Claude Code. | Python nodes that can carry their own HTML, CSS and JavaScript interface, WebAssembly included. Penny writes them on every plan, editing code by hand needs Studio or Enterprise, and on Studio outside AI clients write nodes through Kaitoi's MCP server. Hosted nodes use packages from an approved list, code that needs any other package runs on your machine as a Runtime recipe, and nodes and templates publish to Community. |
| Models Which generative models you can reach, and how they are paid for. | Hosted models from Google, OpenAI, Black Forest Labs, ByteDance, Runway, Luma, Kling, ElevenLabs and others, through Griptape's proxy on credits or on your own API keys, and open image and video models on your own GPU. A first-party library trains FLUX.1 LoRAs locally. | More than 1,400 generative models and around 400 language models as of September 2026, among them OpenAI and Google Gemini models, FLUX, ByteDance's Seedream and Seedance, LTX video models, Runway, Luma's Uni-1, Photon and Ray models, Kling and ElevenLabs, on credits at the provider's own price with no markup or, for providers that accept them, on your own keys. Open models run on your own GPU through your ComfyUI or as Runtime recipes, and a FLUX.2 trainer node, the only trainer, trains LoRA weights from your images as a paid node through fal. |
| Assistant An AI that builds or changes the workflow for you. | A built-in agent adds and connects nodes, runs the workflow and inspects results, asks before it runs workflows, generates images or uses outside tools, follows skill files, and connects to outside MCP servers. Agent nodes call a language model as a step in a workflow. | Penny builds and edits graphs, writes nodes, edits storyboards, timelines and Lab boards, follows Skills and Playbooks, and calls outside MCP servers, on every plan; each reply lists the tools it used, and a stop icon ends the request. Studio shows a credit estimate before a run, and Kaitoi Code, in the terminal, has Plan, Build and Crew modes. Language-model nodes reach around 400 language models. |
| Outside agents and code Driving the product from other AI tools and your own software. | An MCP server lets tools such as Claude Code or Cursor build and run workflows. A Python scripting interface creates and runs flows, pipeline scripts trigger workflows over local HTTP, and the open-source Griptape Framework builds agents and pipelines in code. | Kaitoi's MCP server, on every plan, lets compatible AI clients build graphs, run approved work and inspect results, and on Studio write nodes. A REST API runs graphs from your own code, with no published SDK, and Kaitoi Code, the terminal agent in Kaitoi CLI, can use Kaitoi workflows from a code project. Kaitoi Open Source is announced as coming soon. |
| Running workflows Running all or part of a workflow, and keeping versions. | Run up to or from any node, in parallel or one node at a time, with queued runs and cancel; lock nodes to keep their results; loop over lists with for-each nodes; and save numbered workflow versions. | Run one node, a selection, or every node, with the upstream steps they need, and stop a run while finished steps keep their results. Independent nodes run at the same time, and a run reuses the results of steps that did not change. One output can feed several nodes and an array input collects several results, but there is no node that runs a workflow once per item. The project keeps the graph, with templates and backups to return to an earlier version. |
| Editing and finishing Cutting, compositing and upscaling the result. | Nodes cut, join, re-time and grade video clips, Image Bash composites image layers with mask painting, and a community node lays out storyboards as grid images. Images and video upscale with Topaz Labs, SeedVR2 and local upscalers. | A storyboard of scenes and shots feeds a multi-track timeline that renders MP4, WebM or MOV in ProRes and exports and imports OpenTimelineIO, and colour nodes grade. Video edit, extend and reframe models run as nodes, with image upscalers from SeedVR and others, Topaz and FLUX video upscaling, overlay, blend and mask nodes, and a drawing canvas for masks. There is no layer-stack compositor in one editor. |
| 3D and splats Working with 3D models and Gaussian splats. | 3D assets can be generated from images and loaded into workflows, the editor previews common 3D formats, and it can preview Gaussian splat files. | 3D generation through Tripo, Meshy, Hunyuan3D and World Labs, model previews and a Gaussian splat viewer, a 3D Editor node with render passes and Scene GLB export, and Blender rendering through a connected Blender. |
| Sharing and delivery Handing a workflow to other software or other people. | Publish a workflow as a folder that runs headless, or as a Griptape Cloud Structure run through an HTTP API. Teams share workflows as files or through optional cloud sync. | A saved template becomes an asynchronous API endpoint, polled or streamed and run on your own credits, a REST API runs graphs from your own code, and a Station puts a workflow at a stable link as a simple form or a custom app, where each visitor pays for their own runs; a Station that runs your own code is invite only. Templates, read-only share links with optional cloning, and team workspaces share a graph, and Create subgraph folds a selection into one reusable node. |
| Teams and enterprise Working as a team, and the controls a studio needs. | Professional covers up to three users. Enterprise adds unlimited users, on-premises deployment through an Admin Server, seat management and Cedar permission policies. | Team workspaces with roles, shared projects, pooled credits and weekly caps per member, on a team Studio subscription where every member takes a seat. Kaitoi never trains models on your work. Studio is hosted, and a version you run yourself, Kaitoi Open Source, is announced as coming soon. |
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.
Kaitoi CLI turns an external Hugging Face repository into a runnable local recipe, exposes its parameters inside Kaitoi, and generates a runtime node that joins the graph like any other capability.
This film, from our Kaitoi Signals series, turns an outside model repository into a node that runs on your own machine.
Read more in the documentation:
As of November 2026, Foundry lists a Free plan for testing and non-commercial use, with pay-as-you-go model credits; Professional at $40 per user per month, or $400 per user per year, for up to three users, with $40 of model credits a month; and Enterprise, custom-priced. Local models and your own API keys do not use credits.
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.
Foundry, the maker of Nuke, which completed its acquisition of Griptape on 18 February 2026. Griptape Nodes is still sold on its own, through Foundry's Free, Professional and Enterprise plans.
In part. The engine that runs workflows is open source under Apache 2.0, and so is the Griptape Framework for Python, while the Griptape Nodes application around the engine is proprietary Foundry software. Kaitoi Studio is hosted, and Kaitoi Open Source, a distribution you run yourself, is announced as coming soon.
Not directly. Kaitoi CLI connects your own Blender, EXR nodes read and write OpenEXR images and convert between colour spaces such as ACEScg and Rec.2020, and the timeline hands the cut to DaVinci Resolve or Premiere through OpenTimelineIO, but Kaitoi has no Nuke plugin or connection. Griptape's Nuke library runs Nuke scripts and publishes workflows as Nuke gizmos.
Yes, on every plan. Compatible AI clients can build graphs, run approved work and inspect results through it, and on Studio write nodes. Penny can also call outside MCP servers.