Best C4 model tools in 2026

Architecture diagrams go stale fast. A C4 model is only worth keeping while it matches the systems it describes, and the tool you choose decides how much effort that takes and who gets to see the result.

With AI drafting C4 diagrams from a prompt, drawing is no longer the hard part; choosing a tool whose model you can trust and keep current is. This guide compares five options to help you pick the one that fits how your team works: Eraser, IcePanel, Structurizr, LikeC4, and Mermaid with C4-PlantUML.

What is a C4 model tool?

A C4 model tool keeps a model of your software architecture – people, software systems, containers, components, and the relationships between them – and draws C4 diagrams as views of that model, so each element looks and means the same in every diagram it appears in.

The C4 model describes architecture at four levels of zoom – system context, containers, components, and code – with supporting diagrams for dynamic behavior and deployment. It prescribes the abstractions rather than a notation, which is why the tooling around it varies so much. What is the C4 model? covers the levels and the supporting diagrams in full.

General diagramming tools such as draw.io and Lucidchart offer C4 shape libraries, and Mermaid and C4-PlantUML offer C4 as text. Those give you the notation, but each diagram stands on its own. The tools in this guide are compared on whether there is a model behind the views and how that model is kept current.

Why use a C4 model tool?

• One model, many views. Rename a service once, and every context, container, and deployment view that shows it changes with it.
• A consistent vocabulary. Everyone works from the same people, systems, and containers, with the same names and descriptions, instead of redrawing their own.
• Staying current. Keeping one model in step with the code and infrastructure is cheaper than redrawing a folder of pictures – and some tools can read those sources directly.
• Sharing with people who aren’t architects. Drill-down, flows, and interactive exports let a product owner, an auditor, or a new engineer explore the architecture at the level they need.

The best C4 model tools in 2026

Four of the five keep a model and draw views of it; the fifth draws each diagram from its own text. The comparison below is about that model: what builds it, what happens when a system changes, how people outside the tool get to see it, and where it can run.

C4 model tools compared in November 2026: whether there is one model behind the views, how the model is built, what happens when a system changes, drill-down and flows, interactive HTML export, plain text in git, and deployment options.

Compare	Eraser	IcePanel	Structurizr	LikeC4	Mermaid + C4‑PlantUML
Model	One model, many views	One model, many views	One model, many views	One model, many views	Separate picturesEach file stands alone
Built by	✅Supported: Built-in AI from a prompt ✅Supported: AI from your systems ✅Supported: Text DSL ✅Supported: Drawn on a canvas Systems: code, cloud, CMDB, EA tools	❌Not supported: Built-in AI from a prompt ❌Not supported: AI from your systems ❌Not supported: Text DSL ✅Supported: Drawn on a canvas YAML/JSON import; external agents via MCP	❌Not supported: Built-in AI from a prompt ❌Not supported: AI from your systems ✅Supported: Text DSL ❌Not supported: Drawn on a canvas Java component finder reads code	❌Not supported: Built-in AI from a prompt ❌Not supported: AI from your systems ✅Supported: Text DSL ❌Not supported: Drawn on a canvas Preview edits layout, not elements	❌Not supported: Built-in AI from a prompt ❌Not supported: AI from your systems ✅Supported: Text DSL ❌Not supported: Drawn on a canvas
When a system changes	Rescan updates the mapOn demand, per source	Edit by hand or through the API	Edit the DSLJava components refresh	Edit the DSL	Edit each file
Drill-down and flows	✅Supported: Drill-down ✅Supported: Flows	✅Supported: Drill-down ✅Supported: Flows	✅Supported: Drill-down ✅Supported: Flows	✅Supported: Drill-down ✅Supported: Flows	❌Not supported: Drill-down ✅Supported: Flows Dynamic diagrams
Interactive HTML export	✅ YesOpens in any browser	❌ NoHosted share links instead	✅ Static site	✅ Static siteOr React and web components	❌ NoImages; Mermaid renders in Markdown
Plain text in git	✅ YesDiagrams sync with GitHub and GitLab as .eraserdiagram files	❌ NoExports don’t import back	✅ Yes	✅ Yes	✅ Yes
Deployment options	✅Supported: Multi-tenant SaaS ✅Supported: Single-tenant SaaS ✅Supported: Private cloud ✅Supported: On-premises On-premises includes air-gapped	✅Supported: Multi-tenant SaaS ✅Supported: Single-tenant SaaS ❌Not supported: Private cloud ❌Not supported: On-premises Custom data residency	❌Not supported: Multi-tenant SaaS ❌Not supported: Single-tenant SaaS ✅Supported: Private cloud ✅Supported: On-premises Self-hosted: local and server	❌Not supported: Multi-tenant SaaS ❌Not supported: Single-tenant SaaS ✅Supported: Private cloud ✅Supported: On-premises A static site you host	❌Not supported: Multi-tenant SaaS ❌Not supported: Single-tenant SaaS ✅Supported: Private cloud ✅Supported: On-premises

The comparison reflects each product’s documentation in November 2026: Structurizr v2026.10.26, LikeC4 v1.60.0, IcePanel v112.12, Mermaid 12.3.0, and C4-PlantUML v2.14.0. For Mermaid and C4-PlantUML, flows are their dynamic diagrams.

1. Eraser

Eraser builds C4 models with AI. Describe a system in a prompt, or point Eraser AI at the sources that already describe it, and it builds a map: a file of navigable views over a model, in which each box can stand for a record in that model.

Each C4 level has a natural source, and Eraser reads them:

• Context and landscape. ServiceNow CMDB, Ardoq, and OrbusInfinity, where the application portfolio and its owners already live.
• Containers and deployment. AWS, Azure, Google Cloud, and VMware environments, and infrastructure as code in your repositories.
• Components. Code in GitHub, GitLab, and Azure DevOps repositories.

Maps are built to be read as well as edited. A reader drills down from the system context into containers and components, plays a flow step by step, filters a view by the fields on its records, or presents the views in Present mode. A map exports as an image, a PDF, or an interactive HTML page that anyone can open in a browser. The four levels of a messaging API, a food-delivery platform, and a monolith carved into services are working examples you can explore.

C4 maps live next to Eraser’s other diagrams, so a container view can sit beside the cloud architecture it runs on, the sequence diagram of a critical request, the ERD of its database, and the BPMN process it supports.

Features

• AI-built models. Eraser AI drafts C4 maps from a prompt or from connected data sources; refine the result by prompting or on the canvas.
• Data sources. AWS, Azure, Google Cloud, ServiceNow, VMware, Cisco, Ardoq, and OrbusInfinity, plus CSV, JSON, YAML, XML, and Excel uploads.
• Code as context. GitHub, GitLab, and Azure DevOps repositories inform component views, and Eraserbot watches GitHub and GitLab repositories to keep diagrams current.
• Git sync. Diagrams sync with GitHub and GitLab repositories as .eraserdiagram files, alongside Markdown docs.
• Navigable maps. Drill-down between levels, flows, filters by record fields, and Present mode.
• Rescans. Rescanning a source updates the model and every map built on it.
• Interactive HTML export. A page anyone can open in a browser, alongside image and PDF exports.
• Conversion. Eraser AI converts existing Mermaid, PlantUML, and DOT diagrams.
• Deployment options. Multi-tenant SaaS, single-tenant SaaS, private cloud in your own cloud, and on-premises, including air-gapped.

Best for

• Enterprise architecture and platform teams whose systems are already described in a CMDB, an EA tool, or cloud accounts, and who want C4 views built from those sources rather than redrawn.
• Teams that share architecture with people who will never open an architecture tool.
• Organizations that want C4 alongside the rest of their technical diagrams.
• Regulated enterprises that need their architecture tool single-tenant, in their own cloud, or on-premises.

Limitations

Eraser does not enforce C4 notation. It has no C4 diagram type: C4 lives in maps and in general diagrams – freeform or diagram-as-code – so nothing stops a view from mixing levels, and keeping the conventions is up to you. Sources update when they are rescanned, not continuously, and a rescan updates the model and its maps, not diagrams already drawn on a canvas.

2. IcePanel

IcePanel is a hosted modeling tool built around C4. You draw diagrams by hand, and every object you place lands in a shared model, so a system appears consistently at every level and a double-click zooms into the next one. Its strength is storytelling: flows walk through a scenario step by step, branching into alternate and parallel paths, and tag overlays show different perspectives on the same diagram.

In 2026 IcePanel opened up to agents. Its MCP server reads and writes objects, connections, diagrams, and decision records – diagram create and edit arrived in v112.12 on 3 November. Agent skills turn code or a prompt into C4 objects and diagrams, convert Mermaid C4, and, since 6 October, import Structurizr workspaces. The IcePanel alternative page compares it with Eraser in more detail.

Features

• A model behind every diagram, with drill-down between levels.
• Flows with alternate and parallel paths, and tag overlays.
• Versions, and drafts that merge back into the model as a new version.
• YAML and JSON model import, a REST API, and an MCP server.
• Agent skills for drafting C4 from code or prompts and importing Mermaid C4 and Structurizr.
• PNG and SVG diagrams, a JSON model, CSV, and model summaries as PDF, Markdown, HTML, or LLMs.txt.
• Share links that embed in Confluence and SharePoint.

Best for

• Architecture teams that want a curated, hand-maintained C4 model with polished walkthroughs for stakeholders.
• Organizations where many people read the architecture and few edit it.

Limitations

IcePanel’s model is as current as the last person – or agent – who edited it. Its in-app AI is opt-in and writes descriptions and summaries rather than diagrams, so drafting a diagram goes through an external agent. It no longer links objects to code: repository linking was removed, and IcePanel says an improved version is in the works. Imports add objects but are not a round trip – the documentation says you cannot export, edit the file, and import it back – so the model does not live as text in git. Its HTML export is a summary of the model rather than interactive diagrams; the interactive view is the hosted share link. IcePanel is SaaS only: it offers a single-tenant environment with custom data residency, but there is no private cloud or on-premises option.

3. Structurizr

Structurizr is the reference implementation of C4 as code. You describe a workspace in the Structurizr DSL – people, systems, containers, components, and deployment nodes – and define views over it; every diagram is generated from that one model. It comes from Simon Brown, who created the C4 model, and its tools now ship as one open-source product (v2026.10.26 at the time of writing): a browser playground, local for authoring, server for publishing to your organization, and export.

The viewer is built for C4: double-click an element to open the next level, step through dynamic views and animations, and filter views by tag or switch perspectives. The notation is customizable, with an automatically generated diagram key and prebuilt themes for AWS, Azure, and Google Cloud, and diagrams embed in other pages through an iframe. Deployment views map containers onto infrastructure. For Java systems, a component finder extracts components from compiled code – the closest Structurizr comes to reading your systems directly.

Features

• One DSL workspace, with generated system context, container, component, dynamic, and deployment views.
• Drill-down, animation steps, filtered views, and perspectives in the viewer.
• A Java component finder that extracts components from code.
• Customizable notation, automatic diagram keys, prebuilt AWS, Azure, and Google Cloud themes, and iframe embeds.
• Exports to PlantUML, C4-PlantUML, Mermaid, WebSequenceDiagrams, PNG, SVG, and an interactive static site.
• The DSL lives in git; the server also keeps 30 workspace versions by default.
• AI guidance that has agents write Structurizr DSL rather than diagram markup, and an open-source MCP server that validates, parses, and inspects it.
• Apache-2.0 open source, with prebuilt server binaries that add SAML and role-based access.

Best for

• Teams that want architecture as code, reviewed in pull requests alongside the systems it describes.
• Organizations that need to host their architecture tooling themselves.
• Java teams that can use the component finder.

Limitations

The model is written by hand or by an agent you bring: Structurizr’s AI documentation is a set of guides for external LLMs, and there is no built-in one. Beyond Java components, nothing reads your running systems, so the model drifts unless someone edits the DSL. You run it yourself, in your own cloud or on-premises; there is no hosted version. The server’s open core can be built from source, and the prebuilt server binaries add enterprise features such as SAML and role-based access. The DSL is a barrier for readers who are not engineers, and the Confluence apps are archived, which leaves iframe and image embeds for wikis.

4. LikeC4

LikeC4 is an MIT-licensed take on architecture as code, with about 6,000 GitHub stars and v1.60.0 released on 15 October. As in Structurizr, you write a model in a DSL and define views over it, but the tooling is built around the editor: a VS Code extension shows a live preview as you type.

Views link to one another with navigateTo, so a click on an element opens its more detailed view. Dynamic views can be drawn as diagrams or as classic sequence diagrams, and deployment views map the model onto infrastructure. The CLI builds a static website of the views, ready to host or embed, and generates React and web components for teams that want diagrams inside their own documentation.

Features

• A model DSL with a VS Code extension and live preview.
• Drill-down through navigateTo, dynamic views as diagrams or sequence diagrams, and deployment views.
• PNG, JPEG, JSON, and draw.io exports, and generated Mermaid, DOT, D2, and PlantUML.
• A static website build, plus React and web components.
• An MCP server that gives agents knowledge of the model.

Best for

• Developer teams that want C4 as code with modern editor tooling.
• Teams embedding architecture views in their own documentation site or developer portal.

Limitations

LikeC4 is a set of tools rather than a service, so sharing, access control, and SSO depend on wherever you host the generated site. The model is written and updated by hand or by an agent, and nothing reads your cloud accounts or CMDB. Some newer features – flow-control blocks in dynamic views and parts of deployment views – are marked experimental or in development.

5. Mermaid and C4-PlantUML

Mermaid and C4-PlantUML take the lightest approach: C4 notation as text, one diagram per file. Mermaid supports system context, container, component, dynamic, and deployment diagrams in a syntax compatible with C4-PlantUML, renders in GitHub Markdown, and dropped the “experimental” label from its C4 documentation on 30 October 2026. C4-PlantUML (MIT, v2.14.0) is the more complete of the two, adding system landscape diagrams, sprites, custom tags, legends, and links. The Mermaid alternative page covers Mermaid beyond C4.

Features

• Context, container, component, dynamic, and deployment diagrams in both.
• Plain text that diffs and reviews in pull requests.
• Mermaid renders in GitHub Markdown without extra tooling.
• C4-PlantUML adds sprites, custom tags, legends, and links.
• Both are MIT-licensed open source.

Best for

• Teams with a handful of C4 diagrams that live next to the code in READMEs and docs.
• A first step into C4 before committing to a model-based tool.

Limitations

There is no model. Each file is an independent picture, so a renamed system has to be renamed in every diagram that shows it, and nothing checks that the container diagram agrees with the context diagram. There is no drill-down from one diagram into another. Mermaid’s C4 support is thinner than C4-PlantUML’s: its own documentation lists sprites, tags, links, and the legend as unfinished, and its layout is not fully automatic – the order of statements and a shapes-per-row setting decide where boxes land.

Key features to look for

Model or picture

Decide this first, because it shapes everything else. A picture tool – a shape library, or a Mermaid file – draws C4 notation, but each diagram is its own source of truth. A model tool stores each person, system, and container once and generates every view from it, so the views cannot disagree. Beyond a few diagrams or a single author, you want a model.

Where the model comes from

Someone, or something, has to put the systems into the model. In Structurizr and LikeC4 it is the DSL, written by an engineer or an agent. In IcePanel it is the canvas, imports, and the API. Eraser starts from the systems of record that already describe your estate: a CMDB or EA tool for the landscape, cloud accounts and infrastructure as code for containers and deployment, and repositories for components. Ask which of your existing sources each tool can use, and at which level.

How it stays current

A model that matches last year’s architecture misleads the people who rely on it. In DSL and canvas tools the model changes when someone edits it, so staying current is a process question: who updates it, and when. Structurizr’s component finder refreshes Java components from compiled code. In Eraser, rescanning a data source updates the model and the maps built on it. Scans run when you connect a source and when you rescan it, not on a schedule, so build rescans into your release or review cadence.

Drill-down and dynamic views

C4 is about zoom, so check that a reader can move from a system context diagram into its containers and on into components without hunting for another file. All four model tools do; Mermaid and C4-PlantUML do not. Then check dynamic views – how a request moves through the containers. Structurizr animates them step by step, LikeC4 can draw them as sequence diagrams, IcePanel’s flows branch into alternate and parallel paths, and Eraser’s maps carry flows that play step by step.

Sharing outside the tool

Most people who need the architecture will never have a seat in an architecture tool. IcePanel serves them with hosted share links. Structurizr and LikeC4 export a static website you host yourself. Eraser exports an interactive HTML page that travels as a file – it can sit in a SharePoint library or on a Confluence page and open in any browser. Mermaid and C4-PlantUML produce images, which embed anywhere but cannot be explored.

Git and review

If architecture changes go through pull requests, the model needs to be text. Structurizr, LikeC4, Mermaid, and C4-PlantUML are plain text and diff cleanly, and Eraser diagrams sync with GitHub and GitLab repositories as .eraserdiagram files. IcePanel keeps versions and drafts inside the product instead.

Enterprise needs: deployment, SSO, and Confluence

Banks, insurers, and other regulated organizations usually check three things before anything else:

• Deployment. Many need a single-tenant environment, a deployment in their own cloud, or one on-premises, and the tools split cleanly on it. Eraser offers all four models – multi-tenant SaaS, single-tenant SaaS, private cloud, and on-premises, including air-gapped. IcePanel is SaaS only, with a single-tenant environment as an option. Structurizr, LikeC4, Mermaid, and C4-PlantUML have no hosted service at all: you run them in your own cloud or on-premises.
• Single sign-on. IcePanel supports SAML, Structurizr’s prebuilt server binaries add SAML and role-based access, and Eraser offers SSO. LikeC4, Mermaid, and C4-PlantUML inherit whatever protects the site or repository they live in.
• Confluence and SharePoint. IcePanel share links embed in both. Structurizr’s Confluence apps are archived, leaving iframe and image embeds. Eraser’s HTML export is a file you store alongside the pages that reference it.

Choosing a C4 model tool

Start from your readers and your sources. Who needs to see the architecture, and where will they look at it? Where does the truth about your systems live today – in code, cloud accounts, a CMDB, an EA repository, or people’s heads? A tool that reads the sources you already maintain will stay current with less effort than one that depends on discipline. Then decide how much you need architecture in git, and where the tool has to run. In short:

• Eraser – your systems are already described in a CMDB, an EA tool, cloud accounts, or code, and you want C4 maps built from them, shared as an HTML file, and deployed where your security team needs them.
• IcePanel – you want a hand-curated C4 model with polished flows and share links embedded in Confluence or SharePoint.
• Structurizr – you want the reference implementation of C4 as code, reviewed in pull requests and run on your own infrastructure.
• LikeC4 – you want C4 as code with a live preview in VS Code and diagrams embedded in your own documentation site.
• Mermaid and C4-PlantUML – you need a few C4 diagrams in Markdown next to the code, and no new tool.

Pilot your shortlist on one real system. Model it, then change something – add a container, rename a service – and see how much work it takes for every view to catch up. Then share the result with someone who is not an architect and watch what they do with it.

If you need what is actually deployed rather than a C4 model of it, compare cloud diagram tools. And for diagrams kept as text without a model behind them, see the diagram-as-code tools guide.