The whole studio is
the agent's toolbelt.
NodeTool isn't a chat box bolted onto an editor. Every surface — the node canvas, the sketch pad, the storyboard, the video timeline, the app builder — is exposed to agents as tools, around 120 in all. Hand over a brief and the agent plans the shots, builds the workflow, runs it across Flux, Seedance, Veo, Kling, Suno, and ElevenLabs, and retries what misses. You direct, it executes.
Creative Brief
ID: brief
Art Director Agent
ID: art_director
Image Generation
ID: image
Soundtrack
ID: music
Final Spot
ID: render
Say it. Watch it build. Keep it.
The agent works with the same tools you click. It builds the actual workflow on your canvas, runs it in front of you, and leaves behind something you can rerun — not a chat transcript.
Say what you want
Describe the spot, the app, or the pipeline in chat. The agent plans the steps and authors the graph itself — picks the nodes, wires the edges, selects the models: Flux, Seedance, Veo, Suno, ElevenLabs.
Watch it work
Every tool call lands on your screen: nodes appear, renders come back, the timeline fills. When a render misses the brief the agent retries, and when a call is yours to make it stops and asks.
Keep what it built
The result is a workflow or mini app you can inspect, edit, and rerun for the next campaign — with an agent supervising the failure path on a decision and cost budget you set.
Agent-first, not agent-flavored.
Most tools bolted a chat panel onto an editor. NodeTool went the other way: the entire app is built as tools an agent can operate. Agents don't just answer questions about your work — they do the work, on the same surfaces you use.
The whole app is the toolbelt
Around 120 tools cover every editor: the node canvas, the layered sketch pad, the storyboard, the video timeline, the script editor, the 3D scene, and the app builder. If you can click it, an agent can drive it.
Agents build workflows
Describe the pipeline and the agent authors the graph itself — picks the nodes, wires the edges, selects the models — and validates it before anything runs. What it leaves behind is a workflow you can inspect, edit, and rerun.
Agents build apps — and test them
Ask for a mini app and the agent plans the workflows, places the widgets, wires them together, then runs every interaction and has a second model judge whether the result does what you asked. No passing verdict, no app.
An agent on the failure path
Supervised runs put an agent on call: when a step fails mid-run, it decides — retry, repair the output, skip the item, or stop — inside a decision and cost budget you set, with every intervention logged.
It asks instead of guessing
When a job is missing something only you can decide — a name, a look, permission to delete — the agent stops and asks. Every plan, tool call, prompt, and dollar of model spend is on the record.
Bring your own agent
The full toolbelt is exposed over MCP. Point Claude Desktop, Claude Code, or any MCP-aware agent at NodeTool and it gets the same tools the built-in chat uses: build workflows, run them, read the results.
The newest models, the day they arrive.
The leading image, video, audio, and language models from every major provider, or the same kind of models running on your own computer when you prefer that.
Agents your studio actually ships with.
Open one, add your style guide, point it at the providers you already pay for, and let it run while you direct. Starter workflows live in the examples repository.
Brand Pack Generator
One agent, one brief, a full launch kit. Hero stills, social cuts, ad variants, and a music bed — all in your palette, all on brief. Hand it the style guide once and let it carry the look across the deliverables.
Tell the agent
"Generate the launch pack for the spring drop in our brand palette"
Video Ad Director
An agent that storyboards a 15-second spot, picks Seedance for the hero shot and Kling for B-roll, animates the cuts, and lays a Suno track underneath. You bring the brief, it brings the cut.
Tell the agent
"Direct a 9:16 sneaker ad, neon noir, ending on the logo"
Music Video Workflow
Drop in a track, let the agent pace cuts to the beat, pick scenes that match the lyrics, generate visuals across Flux and Veo, and stitch the final video. You direct, it edits.
Tell the agent
"Cut a moody music video for this Suno track, 4 scenes"
Batch Retouching Agent
Hand it a folder of raw shots. It matches color to your reference, upscales, removes the clutter in the background, and exports the keepers while you get on with something else.
Tell the agent
"Retouch this shoot to match the editorial reference"
Social Calendar Agent
Give it the campaign and the calendar. It plans the posts, generates the visuals, drafts the captions, and queues 30 days of content in your voice across IG, TikTok, and X formats.
Tell the agent
"Plan and generate four weeks of launch posts in our voice"
Variant Explorer
Ten versions of the hero frame in five aspect ratios, ranked by how closely they match the brief. The agent works across providers, scores the results, and shows you only the ones worth your time.
Tell the agent
"Give me 10 alts of the hero shot, ranked by brief match"
Mood Board Researcher
Tell it the vibe — "Wong Kar-wai meets a Tokyo arcade" — and it pulls references, distills a palette, drafts a shot list, and seeds prompts the rest of your canvas can run with.
Tell the agent
"Build a mood board for a perfume launch, dreamy + analog"
Script-to-Storyboard
Paste a script. The agent breaks it into beats, generates a storyboard frame for each, suggests camera moves, and hands the boards to a video model when you give the green light.
Tell the agent
"Storyboard this 60-second narration scene by scene"
Voiceover & Localization
An ElevenLabs agent that drafts read-throughs, clones a presenter for pickups, and dubs the same spot into five languages — lip-sync timing intact, brand voice consistent.
Tell the agent
"Localize this ad into ES, FR, DE, JP, PT — keep my voice"
Every tool in the creative kit.
The same models the studios use, in the hands of agents that know when to reach for them. Hand your agent a brief and it picks the right tool for the shot.
Image models
Video models
Music & voice
Retouch & edit
Brand & references
The model providers
Your favorite model isn't in the list yet?
NodeTool is open source. Wrap any provider as a node, plug in a private checkpoint, or drive the canvas from the TypeScript SDK — agents pick it up like a first-class tool the moment it lands on the canvas. And because the whole toolbelt speaks MCP, outside agents get it too.
Frequently asked questions
What is NodeTool?
NodeTool is the open creative AI workspace. Every major model from every major provider, including FAL, KIE, OpenAI, Anthropic, Gemini, and Replicate, sits on one visual canvas that you run on your own machine or in the browser. Image, video, music, and text share that canvas, alongside agents that carry out longer jobs step by step.
Read more →What does "bring your own keys" mean for me?
It means you connect your own provider accounts and pay each provider directly at their list price. NodeTool never adds a markup, never sells its own credits, and never runs models on its own servers. Your keys, your bill, your data. You will sometimes see this written as BYOK.
Read more →Which models are supported?
The leading models, including Flux, Seedance, Wan, Veo, Kling, Hailuo, Qwen Image, Whisper, ElevenLabs, and Suno, reached through providers such as FAL, KIE, OpenAI, Anthropic, Gemini, Replicate, Together, Groq, Mistral, OpenRouter, and HuggingFace. Models can also run on your own computer through MLX, Ollama, llama.cpp, vLLM, and LM Studio.
Read more →What does it mean that NodeTool is agent-first?
Every editor in NodeTool — the node canvas, sketch pad, storyboard, video timeline, script editor, 3D scene, and app builder — is exposed to agents as tools, around 120 in all. An agent doesn't describe what you could do; it builds the workflow, runs it, and repairs what fails, on the same surfaces you use. The same tools are exposed over MCP, so outside agents such as Claude Desktop and Claude Code can drive NodeTool too.
Read more →What is a planning agent?
A planning agent takes a goal, breaks it into steps, picks a model or tool for each step, and works through them in order, adjusting as it goes. In NodeTool an agent is a block on the canvas, so its results feed into the rest of your workflow.
Read more →Made with working creatives
NodeTool is open source under AGPL-3.0. Star the project on GitHub, join us on Discord, and share workflows with the artists, motion designers, and studios already using it on real jobs.
Get in
touch
Questions, bug reports, feature requests — we answer.