Smart Excalidraw Next: Natural Language to Editable Excalidraw Diagrams

The blank canvas problem Smart Excalidraw Next targets

Drawing tools are fast once a diagram exists and slow before it does. The README frames the project around that gap: you describe the diagram in natural language, and the tool produces a structured, laid-out Excalidraw scene. The stated audience is anyone who needs a flowchart, architecture diagram, sequence diagram, ER diagram or mind map and would rather start from a generated draft than from an empty canvas. The repository lists more than 20 diagram types and says the model can pick a type on its own when the prompt does not specify one. The project is a web application, not a library, so it competes with the act of opening Excalidraw and drawing, not with a rendering engine.

How the prompt becomes an Excalidraw scene

The stack in package.json is Next.js 16, React 19, @excalidraw/excalidraw ^0.18.0, Tailwind CSS 4 and Monaco Editor. The README describes a server-side LLM configuration with a provider type, a base URL, an API key and a model name, which means the browser sends the prompt to a Next.js route that forwards it to OpenAI or Anthropic and returns something the app converts into Excalidraw elements. The README also claims a dedicated arrow-connection algorithm that computes connection points to avoid crossed lines. That claim is not backed by a published specification, so treat the layout quality as something you judge on your own prompts. Monaco Editor is in the dependency list, which suggests the app exposes a raw view of the generated scene, though the README does not document that panel. Because output is Excalidraw format, everything generated stays editable on the canvas: move shapes, restyle them, add elements by hand.

Running Smart Excalidraw Next locally with pnpm

The README gives a four-step local setup. Clone the repository, enter the directory, install with pnpm, and start the development server. The package.json scripts confirm that pnpm dev maps to next dev --webpack and that the project uses pnpm-lock.yaml, so npm or yarn would diverge from the lockfile.

git clone <your-repo-url>
cd smart-excalidraw-next
pnpm install
pnpm dev

The README says the app is then available at http://localhost:3000. For a production build, package.json defines pnpm build (next build --webpack) and pnpm start (next start), though the README does not walk through deployment.

To avoid asking every user for their own key, the README documents a server-side LLM mode driven by environment variables. Copy .env.example and fill in the values.

cp .env.example .env

ACCESS_PASSWORD=your-secure-password
SERVER_LLM_TYPE=anthropic
SERVER_LLM_BASE_URL=https://api.anthropic.com/v1
SERVER_LLM_API_KEY=sk-ant-your-key-here
SERVER_LLM_MODEL=claude-sonnet-4-5-20250929

The .env.example file ships these keys commented out, with openai and https://api.openai.com/v1 as the sample provider values. Restart the dev server after editing. Users then open the access password panel in the top right, enter the password, verify it, tick the enable box and save. The README states that when the access password is enabled, the server-side configuration takes priority and the local one is ignored.

The client-side path is the alternative: open the LLM configuration panel, choose OpenAI or Anthropic, paste an API key, pick a model, and save. The README recommends claude-sonnet-4.5 and states that all configuration stays in the local browser.

Where Smart Excalidraw Next stops being the right tool

The project is a single-user drawing assistant, not a diagram pipeline. There is no documented HTTP API, no CLI, and no import or export command in the README, so you cannot generate diagrams from a script or a CI job. There are no retrieved releases, and package.json reports version 0.1.0, so there is no versioned artifact to pin against. The licence section of the README says MIT License, but the repository metadata returns no licence identifier, and that mismatch is worth resolving before you depend on the terms. The README also documents no rollback, no undo of a server-side configuration change, and no rate limiting or spend cap on the server-side key. If you run the access password mode for a large group, every generation spends your key with no documented ceiling. Finally, diagrams are stored as Excalidraw scenes, so teams that keep diagrams as text in Git get a binary-ish artifact that diffs badly.

Excalidraw with an AI layer versus Mermaid and PlantUML

Mermaid and PlantUML take the opposite approach: you write a text definition, and the renderer draws it. That makes the source reviewable in a pull request and regenerable in CI, which Smart Excalidraw Next does not offer because its output is a canvas scene. The trade is control against speed. Text tools give you deterministic layout and a diffable file; Smart Excalidraw Next gives you a plausible first draft in seconds and then hands you an editable canvas where you fix the details by hand. The README's own framing of the Excalidraw integration is exactly that combination: AI generation plus manual refinement. If your diagrams must be reviewed as code, this is the wrong category of tool. If they are working sketches that live in a canvas and get adjusted visually, the generated draft saves the slow first ten minutes.

Maintenance, upgrade cost and the licence question

The last push to the default branch was on 2026-09-17, and the repository is not archived. The dependency set is the maintenance surface: Next.js 16.0.1, React 19.2.0, @excalidraw/excalidraw ^0.18.0, Monaco Editor and Tailwind CSS 4. Major-version bumps in Next.js or the Excalidraw package are the upgrades most likely to break the canvas integration, and because there are no releases, you upgrade by pulling main. That is the real cost: no changelog, no tagged version, no migration notes. The README states MIT License, while the repository metadata carries no licence identifier. If you plan to redistribute the app or host it for others, confirm the licence text in the repository before you rely on it. This is not legal advice; read the file.

Editorial conclusion

Adopt Smart Excalidraw Next if you already draw in Excalidraw and want a first draft from a sentence rather than a blank canvas, and if you can supply an OpenAI or Anthropic API key or accept a server-side access password. Skip it if you need a documented, versioned API, a pinned release you can audit, or a tool that produces Mermaid or PlantUML text instead of a canvas. Before committing, read README_EN.md, check whether your model name matches what the server expects, and decide whether the access password path fits your team.

Frequently asked questions

Official sources

1. Issues
2. liujuntao123/smart-excalidraw-next on GitHub
3. Project website
4. README