NodeTool
NodeTool is an open-source, agent-first creative workspace: one node-based canvas for image, video, audio and text, where agents drive every editor. You bring your own provider keys and pay model providers directly, without markup.
What is NodeTool?
NodeTool is an open-source creative workspace built around a node-based canvas: each node performs one typed operation and the edges between them carry the data. The graph is the document — a file you keep, re-run on new inputs, hand to a colleague or call from code — rather than a chat transcript that vanishes once the answer arrives. Image, video, audio and text share that single canvas, with the editing tools sitting next to the models instead of in a separate application.
What separates it from a canvas with a chatbot bolted on is that it is agent-first. The node canvas, the sketch pad, the storyboard, the video timeline, the script editor, the 3D scene and the app builder are all exposed to agents as tools, around 120 in total. Describe the piece you want and the agent plans the steps, picks the nodes, wires them together, chooses the models, runs the workflow in front of you and retries what misses; when a decision is genuinely yours, it stops and asks. The same toolbelt is published over MCP, so Claude Desktop, Claude Code or any MCP-aware agent can drive the workspace from outside.
The economics are the second distinguishing trait. NodeTool sells no credits and runs no models on its own servers: you connect your own API keys to FAL, KIE, OpenAI, Anthropic, Gemini, Replicate and others, and each provider bills you its list price with no markup. A built-in dashboard records what each step spends and totals it by workflow, provider or model, exportable as a spreadsheet.
Two editions share one AGPL-3.0 codebase. NodeTool Studio is the free desktop application for macOS, Windows and Linux; it needs no account, keeps workflows, files and keys on your disk, and runs open-weight models locally through MLX, Ollama, llama.cpp, vLLM and LM Studio, offline once they are downloaded. NodeTool Cloud is the same application hosted in the browser, currently in alpha and explicitly not generally available. Either way, the code can be self-hosted at any time.
What it does
- Describe a brief in chat and let an agent build the workflow and run it
- Generate image, video, audio and text on a single node-based canvas
- Edit where you generate: mask, inpaint, outpaint, relight, upscale and composite
- Cut a multi-track video and audio timeline and export a finished sequence
- Run open-weight models on your own machine and keep working offline
- Track what every run actually costs, by workflow, provider or model
- Drive the whole workspace from Claude Code or any other MCP-aware agent
When to use NodeTool / When not to
A quick filter to help you decide if NodeTool is the right fit.
When to use NodeTool
- Creative professionals who currently juggle separate tools for stills, motion, voice and the final edit
- Marketing teams producing campaign assets at volume, one variant per product, market or audience
- Studios and freelancers who want their files, workflows and API keys to stay on their own machine
- Developers who want to call creative workflows from TypeScript, a CLI or their own custom nodes
- Researchers summarising document corpora and extracting structured data with local models
When not to use NodeTool
- Teams that need a guaranteed service level today, since the hosted Cloud edition is still in alpha
- Anyone unwilling to open, fund and manage API accounts at each model provider
- Mobile-first users, as there is no iOS or Android application
- Organisations that require a formal SOC 2 or ISO 27001 certification before adopting a tool
- People who want one perfectly polished asset rather than a repeatable production pipeline
How to use NodeTool
A typical end-to-end flow, from setup to results.
- Download NodeTool Studio from the GitHub releases page for macOS, Windows or Linux, with no account required
- Or open the Cloud alpha in your browser at app.nodetool.ai if you would rather install nothing
- Add your own API keys for the providers you use, such as FAL, KIE, OpenAI, Anthropic, Gemini and Replicate
- Optionally install local models through Ollama, MLX or llama.cpp to keep working without a network connection
- Open one of the workflow templates, or start from an empty canvas
- Drag in model, edit and file nodes and connect them, or describe the result in chat and let the agent author the graph
- Run the workflow and watch each step render, inspecting frames as they arrive
- Swap a model or rewrite a prompt and re-run from that point rather than from the beginning
- Finish inside the built-in editors: mask and retouch on the sketch canvas, trim and arrange on the video timeline, then export
- For code, use the TypeScript SDK and graph DSL, write custom nodes, or run the workflow from the CLI
Pros & Cons
Pros
- No markup on model calls: a run billed 0.18 USD at the provider costs 0.18 USD in NodeTool
- Entirely AGPL-3.0, with nothing held back for a paid tier and self-hosting possible at any time
- No lock-in: switch model or provider in one click, and compare the same model across FAL, Replicate and KIE
- Studio works fully offline with local models, keeping workflows, files and keys on your own disk
- One application covers what usually takes four or five: stills, motion, voice and the final edit
- Every agent plan, tool call, prompt and unit of model spend is logged and inspectable
- The site documents its own limits instead of hiding them
Cons
- The Cloud edition is in alpha and explicitly not generally available, with breaking changes and downtime announced
- No price is published for the Cloud subscription: it will be set at full release
- Local models demand real hardware, around 16 GB of RAM, 4 GB of graphics memory and 20 GB of disk
- Bringing your own keys shifts the bill to the providers, it does not remove it
- The @nodetool-ai packages are not on npm yet, so the SDK runs from a source checkout
- No mobile application, and Studio updates have to be installed manually
- No security certification is published and no interface language other than English is announced
Pricing & Plans
A permanent free plan exists: NodeTool Studio, the desktop edition, is free forever under the AGPL-3.0 licence. NodeTool Cloud is a subscription covering hosting, but no price is published, the vendor stating that pricing will be set at full release; the edition is free during the current alpha. In both editions, model usage is billed by the providers themselves at their list prices, with no credits and no markup, while models run locally carry no per-call cost.
- desktop application for macOS
- Windows and Linux
- local models through Ollama
- MLX and llama.cpp
- your own keys for every cloud provider
- workflows and files kept on your machine
- updates installed by you
- works fully offline
- the same AGPL-3.0 workspace in the browser
- nothing to install
- no graphics card needed
- your own keys stored encrypted
- managed storage encrypted at rest
- always on the latest version
- no local models and an internet connection required
Data, GDPR & hosting
A consolidated view of how NodeTool handles your data.
GDPR overview
Compliance is claimed explicitly rather than implied: the privacy policy, last updated on 2 May 2026, states that NodeTool follows the EU General Data Protection Regulation and the German Federal Data Protection Act. Concrete measures are named. The Amsterdam provider publishes an imprint under Art. 13 GDPR, a Data Protection Officer reachable at dpo@nodetool.ai and a Digital Services Act contact point. Rights under Articles 15 to 21 and withdrawal of consent under Art. 7(3) are listed, together with the right to complain under Art. 77 to the Dutch Autoriteit Persoonsgegevens. Server logs rest on legitimate interest, analytics run cookieless through Plausible, and transfers outside the EEA rely on an adequacy decision or standard contractual clauses. One inconsistency stands out: a Dutch company invoking German data protection law and German jurisdiction.
Who owns the data?
The Terms of Use state that workflows, prompts, files and outputs created in the desktop application stay on the user's device, and that the user retains all rights to that content. Ownership therefore never transfers to the vendor. The user remains responsible for holding the rights to any input processed and for complying with the terms of the third-party model providers they connect. In the hosted Cloud edition, workflows and files sit on the vendor's managed servers, encrypted at rest, but the same ownership rule applies. The NodeTool name, logo and branding remain with the NodeTool team: beyond the rights granted by the AGPL-3.0 licence, no further rights are conceded.
Reuse rights
Under the Terms of Use, the end user may reuse what they produce without asking permission: they retain all rights to their workflows, prompts, files and outputs. Two conditions frame that freedom. First, the user must genuinely hold the rights to whatever they feed into a workflow. Second, every hosted model called from the canvas is governed by the terms and privacy policy of the provider that runs it, so the commercial reuse of a generated image, clip or track depends on that provider's licence rather than on NodeTool. The Terms also disclaim any warranty over model outputs, note that they can be inaccurate or biased, and place the duty of review on the user before relying on them, particularly in medical, legal, financial or safety-critical work.
Data retention & training
Hosting summary
Hosting depends on the edition. NodeTool Studio hosts nothing: workflows, prompts, files, generated outputs and API keys stay on the user's own machine, and the privacy policy states that they are neither collected nor synchronised. NodeTool Cloud stores workflows and files on the vendor's managed servers, encrypted at rest. For the website and the hosted service, the privacy policy of 2 May 2026 states that infrastructure runs on EU-based providers and that persistent application data is stored in Frankfurt, Germany. Static content may be served from a global content delivery network, where only the technical data described for server logs is processed, on the basis of standard contractual clauses where applicable. Where a processor sits outside the EEA, the vendor relies on an adequacy decision or on standard contractual clauses with supplementary measures. The named processors are a hosting and CDN provider, Plausible Analytics (Plausible Insights OU, Estonia), email providers and GitHub. A DNS lookup on 11 August 2026 resolved the domain to a Cloudflare anycast address geolocated in the United States, which reflects the edge network rather than where data is stored.
Things to keep in mind
Risks and trade-offs to weigh before adopting NodeTool.
- The hosted Cloud edition is in alpha and the vendor itself announces breaking changes, rough edges and downtime, so it is not yet a safe place for client deadlines
- Bringing your own keys makes spending invisible until the provider invoice arrives: the built-in cost dashboard exists for a reason, and an expensive video model can be run dozens of times without anyone noticing
- An agent that builds and runs workflows on its own can erode the habit of understanding the pipeline, yet when a step fails someone still has to know why
- Generated images, clips and voices are governed by each provider's licence rather than by NodeTool, so commercial reuse can be restricted even though the workspace itself is permissive
- AGPL-3.0 is a reciprocal licence: embedding the code in a closed product carries obligations that should be checked before any integration
- The legal documents are inconsistent, with a Dutch company invoking German data protection law and German jurisdiction, and the data controller named as the NodeTool team rather than the registered legal entity
- No security certification is published and the Cloud edition stores workflows and files on the vendor's servers, so sensitive material is safer in Studio, where nothing leaves the machine
Setup & Integrations
Technical difficulty
Low to moderate, depending on the edition. Cloud takes about thirty seconds: sign in and run, with nothing to install and no graphics card needed. Studio is a signed and notarised installer for macOS and Windows, requires no account, and stays straightforward until local models enter the picture, since those want 16 GB of RAM, at least 4 GB of graphics memory and around 20 GB of disk. Building on the canvas needs no code at all. The developer route is harder: Node 22, a source checkout and a build, as the packages are not published on npm yet.
Deployment
Integrations
Behind NodeTool
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement NodeTool.
Frequently asked questions
Is NodeTool really open source?
How much does NodeTool cost?
What does bringing your own keys mean in practice?
Do I need a graphics card?
Is my data used to train models?
What is the difference between Studio and Cloud?
Can another AI agent drive NodeTool?
Can I call a workflow from code?
Is there a minimum age?
Where is my data hosted?
Should you pick NodeTool?
NodeTool is one of the rare AI tools whose claims can be checked line by line: the source is public under AGPL-3.0, the releases are downloadable, the documentation is live, and a Dutch company with a chamber of commerce number and a VAT identifier stands behind it. The proposition is coherent, namely putting the whole creative chain on one canvas, letting agents operate it through the same actions a human clicks, and taking no cut on the models. For anyone currently paying a markup on credits across three or four separate tools, that is a real argument.
Maturity is uneven, and the site says so itself. Studio is the solid edition: free, offline-capable, running on macOS, Windows and Linux. Cloud is explicitly in alpha and not generally available, carries no published price, and the developer packages have not reached npm. Production work should start with Studio and treat Cloud as a preview.
The cost model deserves a second look before committing. Bringing your own keys removes the markup, not the bill: funded accounts are still needed at each provider, and the built-in cost dashboard exists precisely because those amounts add up. Local models are the genuinely free path, and they ask for hardware in return.
Two points warrant human verification. The legal documents combine a Dutch company with German data protection law and German jurisdiction, an inconsistency the site never explains. And no security certification is published, which will matter to organisations that require one. Neither undermines the product, but both belong in a procurement conversation.
- Choosing a selection results in a full page refresh.
- Opens in a new window.