See it as a diagram
Everything below, as a diagram you can edit. Describe yours and see it in seconds.
No account needed · Editable canvas, not a picture
The three places an AWS diagram can come from
Your hands — manual editors where you place each service and draw each arrow. Total control, total maintenance burden. Your account — scanners that call the AWS APIs and render what is deployed right now. Always true, rarely readable. Your words or your repo — generators that produce the diagram from a description or from infrastructure code, which you then edit. Fast and readable, but only as complete as what you told them.
None of these is best in general; each is best against a different failure mode. The comparison below is ordered from most manual to most automated.
1. diagrams.net (draw.io) — the free manual standard
Free, runs in the browser or offline, and ships the complete official AWS shape libraries. For a one-off diagram where you know exactly what you want to show, it remains the default answer, and nothing about it locks you in — files are XML you can keep in the repo.
The honest limit
2. Lucidchart — the enterprise manual editor
The polished commercial version of the same idea: AWS shapes, team collaboration, comments, SSO, and templates, with paid seats. Where draw.io is a file, Lucidchart is a system of record with permissions — which is what larger organizations actually buy it for. Its cloud-visualization add-on can also import from a live account into an editable canvas.
The maintenance economics stay manual: a human still owns every update, now at a per-seat price. If your organization already pays for it, the marginal cost is zero; if not, the money question is whether you need the collaboration layer or just the diagram.
3. Cloudcraft and Hava — draw what actually runs
Account scanners connect to AWS with a read-only role and render your real deployed infrastructure — Cloudcraft (now part of Datadog) with its signature 3D isometric view and live cost overlays, Hava with automated versioned snapshots per VPC. For an as-built audit, a security review, or the moment someone asks “what are we actually running in eu-west-1”, this category is unbeatable, because the diagram cannot lie.
The honest limit
4. Python diagrams and Mermaid — the repo-native options
The diagrams Python library defines AWS diagrams in code with official icons; Mermaid's architecture syntax renders directly in GitHub READMEs. Both versions of diagrams-as-code share the same virtue — the diagram diffs in review and changes in the same commit as the infrastructure — which is why they pair naturally with Terraform repos.
# diagrams (mingrammer) — the diagram is code
with Diagram("Ingest", show=False):
ALB("alb") >> ECS("api") >> RDS("orders")The cost is layout control: past a couple dozen nodes you are fighting the auto-layout, and there are no zones, no deliberate placement, and no editing by anyone who does not write code. Strong for small per-service diagrams inside repos; weak as the big picture you present.
5. AI generators — describe it, then edit it
The newest category generates the diagram from a plain-English description and gives you an editable canvas to refine. Datadef's AWS generator produces architectures with the real AWS service icons — the same 439-icon library you would use manually — plus zones, labeled flows, and data lineage where it applies. Eraser takes a related approach from the diagram-as-code side with AI on top.
Two properties separate this category from a ChatGPT-drawn Mermaid sketch. The output is a real editable diagram, not frozen syntax — generation gets you to 80% in a minute and the mouse gets the last 20%. And the diagram can live on: embedded by link in a README or Confluence page, it updates everywhere when the canvas changes, which is the property the docs-sync comparison shows almost nothing else has.
The honest limit
6. Agents via MCP — the diagram your AI maintains
If your team already codes with Claude Code, Cursor, or another agent, the drawing step can disappear entirely. Datadef exposes the canvas as an MCP server: the agent reads your Terraform or CDK, draws the AWS architecture with proper icons, and — the part that changes the economics — updates the same diagram in a later session when the infrastructure changes. The agent that made the change documents the change.
Combined with live embeds, this closes the full loop that keeps AWS docs honest: repo changes, agent updates canvas, every embedded copy is current. No other category in this comparison closes that loop today.
The comparison, by where the truth lives
| Tool | Source of truth | Keeping it current | Best for |
|---|---|---|---|
| diagrams.net (draw.io) | Your hands | Manual, every time | Free one-off diagrams |
| Lucidchart | Your hands, shared canvas | Manual, with owners | Enterprise collaboration |
| Cloudcraft / Hava | Your AWS account | Automatic rescan | As-built audits, cost views |
| Python diagrams / Mermaid | Code in the repo | Same-commit edits | Small repo-native diagrams |
| Datadef generator | Your description | Edit canvas, embeds follow | Explainable architecture, fast |
| Datadef + agent (MCP) | Your repo, read by an agent | Agent updates in-session | Teams coding with AI agents |
Choosing by failure mode
“I need a diagram for tomorrow's review” — generate it, then spend your hour editing instead of placing boxes. “I do not trust that our docs match production” — that is an as-built problem; scan the account with Cloudcraft or Hava. “Our diagram is always three quarters behind” — the update loop is broken, not the drawing tool; embed a live diagram and, if you work with agents, let the agent own the updates. “We just need boxes in a README” — Mermaid, free, in the commit.
Most teams land on two artifacts: an as-built view straight from the account when auditing, and one explainable architecture diagram that is generated, edited, and embedded everywhere people actually look.