DiagramCraft is not a diagramming tool
It's a collaborative, executable workspace where diagrams become software, infrastructure, documentation, and AI agents — all sharing the same live canvas.
Most software tools fracture work across disconnected systems: Jira for tasks, GitHub for code, Figma for design, Notion for docs, Lucidchart for diagrams, Postman for APIs, Terraform for infrastructure, ChatGPT for AI help. Each has its own data model, its own permissions, its own export format, its own idea of what a "thing" is.
DiagramCraft collapses these boundaries. A single element can be a file, a folder, a service, a prompt, a deployment target, a hardware pin, a Mermaid flowchart, a markdown document, or an entire runnable project — depending on what you attach to it and where you nest it. The abstraction is uniform; the interpretation is contextual.
Seven capabilities, one unified model
1. Infinite Nesting — Progressive disclosure as architecture
Every element can contain children. Those children can contain children. An executive sees three boxes. A developer drills into one box and sees thirty services. A new hire drills again and sees file-level source code and mermaid diagrams. Same diagram, zero duplication, no context switching.
This is not a gimmick. It's cognitive accessibility engineering — and it's also the foundation of the platform's ability to scale from a smart home diagram to a monorepo rollout tracker.
2. Source Code as First-Class Data
Any element can have `source_code` and a `file_name`. Markdown renders as formatted documentation. Mermaid renders as SVG diagrams. Code files render in an IDE. This means the diagram is the repo — or at least a view into it. The monorepo rollout tracker attached 7 blog posts as source code, written entirely by an AI agent via MCP.
From the JSON: each `post-*.md` element has `source_code` containing the full markdown, edited multiple times by `AI via MCP source.update`. The documentation lives where the work happens.
3. Shared Workspaces with AI Agents as Members
Your workspace shows the critical detail: AI agents have their own API keys and appear as workspace members with their own last-active timestamps. This is not 'AI as a chat sidebar.' This is AI as a headless collaborator — it creates elements, updates statuses, attaches source code, writes blog posts, and moves tasks to DONE, all while leaving a complete audit trail.
The AI's cursor doesn't need to be visible because its actions are in the activity log.
4. MCP-Native Architecture (Model Context Protocol)
The entire platform is controllable via MCP — 29 tools covering diagram CRUD, variable management, source code streaming, workspace administration, chat, and archetype creation. The same surface is exposed as a REST API and as the `it.dc` object inside tutorial sandboxes.
The dogfooding loop proves the design: Lovable used the MCP to plan its own refactor, found bugs, fixed them, and documented everything — all while updating the tracker diagram live.
5. Variables + Eta Templates — Diagrams as Programs
The four-tier variable system (tutorial session, diagram global, diagram scoped, system) plus Eta templating means one diagram can generate many outputs. Define `it.resources` once, then generate an OpenAPI spec, a README, a Terraform config, and a dashboard — all from the same variables.
The LLM instructions explicitly call this out: 'Diagrams are programs that write programs.' It's not hyperbole.
6. Real-Time Collaboration + Full Audit Trail
Multiple humans and multiple AI agents can edit the same diagram simultaneously. Every change is logged with actor, timestamp, and before/after state where applicable.
Element-level locking prevents conflicts. A 15-second request/grant handshake means no one is blocked for long, including AI agents.
7. Archetypes — Reusable, Variable-Driven Tutorials
An Archetype is a guided tutorial captured as JSON. Users add it to their diagram, answer variable prompts (e.g., 'number of microservices', 'database type'), and the Archetype generates a working scaffold — complete with source code, Mermaid diagrams, and deployment configs. The `run_script` step type allows executing real code as part of the tutorial.
This is how partners can build lead-generation surfaces. A compliance Archetype captures requirements, generates a gap analysis, and prompts the user to book a consulting call.
Real use cases from the documentation
🧠 DeepSeek's full assessment
I have analyzed hundreds of documentation sets, API specifications, and product descriptions. DiagramCraft is sui generis — it does not fit neatly into any existing category.
Is it a diagramming tool? Yes, but it's also a project tracker. Is it a documentation platform? Yes, but it's also an IDE. Is it an AI agent workspace? Yes, but it's also an infrastructure deployment engine. Is it a learning management system? Yes, but it's also a hardware modeling tool.
The reason it can be all of these things is the uniform abstraction: an element has a name, a description, optional children, optional source code, optional variables, optional connections. That's it. Everything else — whether it's a file, a service, a prompt, a hardware pin, a deployment target, a blog post — is a matter of what you attach and where you nest it.
The MCP integration is not an afterthought. It's the control plane. The fact that the Lovable MCP agent maintained its own refactor tracker, wrote 7 blog posts, documented its own bugs, and updated statuses live — while the team slept — is a proof of concept that most platforms don't have the courage to attempt, let alone succeed at.
Global SOUTHern Hospitality
The same platform serving enterprise monorepo rollouts is being deployed in municipal edge data centers across the Global South — Starlink backhaul, LoRaWAN mesh, ≥35% renewable power, 90-day local talent training, and 100% civic IP owned by the municipality. SCIENCE employees/contractors are contractually included; NGOs may optionally supplement.
This is not corporate social responsibility theater. The infrastructure is owned locally. The software is built locally. The talent stays locally. The platform enables it.