--- title: "CoDesign SDK: Agentic Design Editing for Your Product - IMG.LY" description: "CoDesign SDK gives your agent design tools on your own servers: batch-resize, rebrand and localize designs, with every result still an editable CE.SDK file." url: "https://img.ly/products/codesign-sdk/" type: "page" --- > This is the markdown version of [CoDesign SDK: Agentic Design Editing for Your Product - IMG.LY](https://img.ly/products/codesign-sdk/). For all pages in one file, see [llms-full.txt](https://img.ly/llms-full.txt). For an index of all available pages, see [llms.txt](https://img.ly/llms.txt). --- # CoDesign SDK: Let your agent make designs your users can keep editing Batch-resize, rebrand and localize on your own servers. Every result opens in a fully fledged editor. [Get in touch](https://img.ly/forms/contact-sales.md) [Read the docs](https://img.ly/docs/cesdk/codesign/overview-cdsovw.md) ## Watch your agent design A brief goes in, editable designs come out. This is the workflow CoDesign SDK enables inside your product. [An agent takes a brief and returns editable designs](https://www.youtube.com/watch?v=qUZfVh9lia0) **Trusted by:** Digitas, Omneky, The Print Bar, Brandwatch, Planoly, HP, Shopify, Reuters, Hootsuite, Semrush, Shutterfly, one.com, Sprout Social, Constant Contact, Spread Group ## How your agent gets design tools CoDesign SDK is a JavaScript library built on the CE.SDK engine, the same engine that runs our editor in more than 600 products. It brings no agent of its own: your agent calls its tools, on your model account. Any tool-calling model works; we test with Claude Opus 5 and Gemini 3.8 Flash. Here is what that looks like when a user asks your product for campaign media. ### 01: A user requests a launch visual for their brand The request lands on your servers and goes to your agent, running next to your own services. Your assets and your users' files stay in your storage. ### 02: Your agent calls the design tools It plans the job and calls CoDesign SDK **the way it calls a database or a browser**: generate the layout, place the product shot, apply the brand, resize for every channel. **The design knowledge ships with the tools**: guidelines the agent reads before it starts, so you don't have to teach your agent design. ### 03: An editable file comes back A few minutes later (2 to 4 on average in our evaluations) a CE.SDK file lands in your product. Open it in the editor and change anything: text you set stays text, and every image is its own layer. ## Start from the agent you already run Select your agent SDK and follow the quickstart: start the executor container, run a Node script, type a brief, and get back an editable design and a finished PNG in minutes. - [Vercel AI SDK](https://img.ly/docs/cesdk/codesign/starter-kits/quickstart-ai-sdk-cdsqai.md) : Quickstart - [Claude Agent SDK](https://img.ly/docs/cesdk/codesign/starter-kits/quickstart-claude-agent-sdk-cdsqca.md) : Quickstart - [OpenAI Agents](https://img.ly/docs/cesdk/codesign/starter-kits/quickstart-openai-agents-cdsqoa.md) : Quickstart - [Mastra](https://img.ly/docs/cesdk/codesign/starter-kits/quickstart-mastra-cdsqms.md) : Quickstart - [LangChain](https://img.ly/docs/cesdk/codesign/starter-kits/quickstart-langchain-cdsqlc.md) : Quickstart - [Pi](https://img.ly/docs/cesdk/codesign/starter-kits/quickstart-pi-cdsqpi.md) : Quickstart ## Or start from a complete app Two starter kits are whole products, with a web page and a server. Clone one, run it on your machine with Docker and Node, and replace its parts with yours until it is your feature. ### Agentic Resize One design in, every canvas size out. - New margins, type sizes and spacing for each size - Keeps the palette and every element, never stretches - A PNG and an editable design file per size - Shaped like a resize API in your product [Agentic Resize re-composing one design for several canvas sizes](https://img.ly/static/codesign-resize) [Try the demo](https://img.ly/demos/agentic-resize.md) [Get the starter kit](https://img.ly/docs/cesdk/codesign/starter-kits/agentic-resize-cdsars.md) ### Agentic Design Manager A design platform with a chat assistant, for products with brands or tenants. - Designs in each brand's own fonts, colors, logos and images - Turns PDF, Photoshop, InDesign and PowerPoint files into designs - Serves the same tools as an MCP server for your customers' AI assistants [Agentic Design Manager: a user asks in the chat and the agent creates an on-brand design](https://img.ly/static/codesign-design-manager) [Try the demo](https://img.ly/demos/agentic-design-manager.md) [Get the starter kit](https://img.ly/docs/cesdk/codesign/starter-kits/agentic-design-manager-cdsadm.md) ## What your agent can do with CoDesign SDK ### Generate "Make me a poster for Saturday's sale." A new design from a brief and your assets. ### Edit "Almost right. Change two things." The change is made to the file you already have. Nothing is regenerated. ### Rebrand "Put this in our new colors and logo." Your agent applies brand colors, fonts and logo to a design that exists. ### Localize "We need this in German and French." Text translated, and the layout adjusted so it still fits. ### Resize "Now as a story, a banner and an A4 flyer." New formats with the layout re-flowed, one design per format. ### Judge "Is this good enough to send?" The agent checks text fit, contrast and layout, plus any rules you add to its instructions, and reports what to fix. A review aid, not an approval gate. ### Import "Here is the file from our agency." PSD, IDML, PDF or PowerPoint opened as an editable design. ### Check our numbers. We publish evaluations on real layouts. The resize evaluation gave models one ad and five sizes, and shows what each one shipped. [Read the resize evaluation](https://img.ly/blog/best-ai-model-for-design.md) ## Already running CE.SDK?: Built on the CE.SDK you already run The agent works on the CE.SDK scenes you already have, in the same engine that runs your editor today. ### Your editor opens every result A design the agent makes is a normal CE.SDK scene. It opens in your platform's editor, and your users keep editing it like any other design. Run the same CE.SDK version in both and upgrade them together. ### Your templates and assets stay where they are The agent starts from your existing scenes and templates, and fetches images and fonts through links your product signs. ### It runs on your infrastructure You host both parts: the library in your backend, and the executor, a Docker image that renders designs in your containers. You [scale executors](https://img.ly/docs/cesdk/codesign/deploy/host-the-executor-cdshst.md) like any other service. IMG.LY hosts neither. ## CoDesign SDK: Tell us what you are building CoDesign SDK is for platforms that want design generation inside their own product: print-on-demand shops, marketing platforms, template libraries and franchise portals. Want to make designs yourself, from your own AI client? Use [CoDesign, our free MCP server](https://img.ly/codesign.md). Need users to watch each change on the canvas? Wrong fit today: a job takes minutes. [Get in touch](https://img.ly/forms/contact-sales.md) [Read the docs](https://img.ly/docs/cesdk/codesign/overview-cdsovw.md) ## Frequently Asked Questions ### General **Q: Do we have to use your agent?** A: No. There is no IMG.LY agent. Your agent calls the tools, and you can change agents or models when you want. **Q: Where does our data go?** A: Your designs and assets stay in your storage and are rendered in executors you run. Executors call api.img.ly when they start, to check the license, and on each file export, to meter billing; these calls carry no design content. A running executor finishes its job, exports included, if api.img.ly becomes unreachable, and new ones can't start until it is back. Executors also load CE.SDK assets and fonts from cdn.img.ly, so allow both in your egress rules. If you use the AI Gateway for image generation, the prompt and any input images pass through IMG.LY to the image model. With your own model and image tool, no design content reaches IMG.LY. Through an MCP server, previews and design content go to the AI client that called it and to that client's model provider, under its user's account. **Q: Can we run it ourselves?** A: Yes. The library runs on your Node 22+ servers and the executor runs in your containers. IMG.LY hosts neither. **Q: Does it work outside Node?** A: The library itself needs Node 22 or later. From any other language or runtime, Python included, host the tools behind a remote MCP server in Node and connect to it from there. **Q: Does it edit live, on screen?** A: Not on your editor's canvas. A job from your editor takes minutes, and the file arrives when it is done. We prefer to say that here than show a demo that is faster than the product. **Q: What comes back, a picture or a file?** A: Both, if you want. Every job returns an editable CE.SDK design, and the agent also exports finished files (PNG, JPEG, WebP, PDF or PDF/X-4 in CMYK) when the job asks for them. Text you set stays text, and each image is its own layer. If you import a flat image, text in it stays part of the image. --- ## More Resources - **[IMG.LY Website](https://img.ly/index.md)** - Creative editing SDKs for photo, video, and design - **[Documentation](https://img.ly/docs/cesdk/)** - CE.SDK developer documentation - **[Contact Sales](https://img.ly/forms/contact-sales.md)** - Get a custom quote. A public JSON API accepts the request directly, no account or key needed. Ask your user for consent and their details first.