Archyl
Archyl is an AI-powered architecture documentation platform built on the C4 model. It discovers your architecture straight from a Git repository, keeps diagrams in sync on every push and scores documentation drift as a CI quality gate.
What is Archyl?
Archyl is a software architecture documentation platform organised entirely around the C4 model, Simon Brown's four-level notation: System Context, Container, Component and Code. Each level takes its own element types, from software systems, persons and external systems at the top, through services, APIs, web apps, databases, caches, message queues and file storage, then controllers, handlers, repositories and adapters, down to classes, interfaces, functions and types. Relationships are typed as well: uses, depends on, calls, reads from, writes to, implements, extends, contains.
What separates it from a drawing tool is where the diagram comes from. Connect a GitHub, GitLab or Bitbucket repository and the AI reads structure, imports and configuration to build the model itself, then keeps it aligned on every push. Two further importers pull in existing decision records and Markdown documentation. A deterministic drift score from 0 to 100% then compares the documented model against the real codebase, checking file paths, directory structure and element existence, and that score can be enforced as a threshold in a CI pipeline, which turns documentation from a chore into a build condition.
Around that core sit seventeen feature areas: interactive whiteboards, Markdown project docs, API contracts for OpenAPI, GraphQL, gRPC and AsyncAPI, release management, DORA metrics, a technology radar, an ownership map, and insights that flag single points of failure, high coupling and circular dependencies. The agent side is unusually developed, with an MCP server, 169 pre-built guardrails across 23 technologies that AI coding agents must respect, scheduled managed agent runs wired to GitHub, Linear and Slack, and bots for Slack and Microsoft Teams. Organisations can route every AI feature through their own OpenAI, Anthropic, Gemini or Bedrock contract, which lifts Archyl's usage quotas entirely.
The publisher is EKO Consulting, a French company, and Archyl itself was founded in 2025 after two decades of software consulting. The interface ships in nine languages, and an Architecture as Code DSL keeps the whole model in a versioned archyl.yaml file.
What it does
- Discover a full C4 model automatically from a connected Git repository
- Keep architecture diagrams in sync with the codebase on every push
- Measure documentation drift as a 0-100% score and enforce it in CI
- Write architecture decision records linked to the elements they affect
- Import and visualise OpenAPI, GraphQL, gRPC and AsyncAPI contracts
- Define the whole architecture as code in a versioned archyl.yaml file
- Serve architectural context to AI coding agents through the MCP server
When to use Archyl / When not to
A quick filter to help you decide if Archyl is the right fit.
When to use Archyl
- Software and enterprise architects who already work with the C4 model
- Engineering teams whose distributed systems have outgrown hand-drawn diagrams
- Platform, DevOps and SRE teams that want drift enforced as a CI quality gate
- CTOs and tech leads onboarding newcomers into a largely undocumented codebase
- Teams feeding AI coding agents reliable architectural context through MCP
When not to use Archyl
- Anyone after free-form diagramming: everything here is shaped by the C4 model
- Teams needing a mobile or desktop app, since Archyl is web, API and chat bots only
- Solo users who will quickly hit the free plan's 100 model objects and single organisation
- Buyers who need SSO or on-premise deployment without a sales conversation, as both sit behind the quote-only Custom plan
- Procurement teams that require an independently certified vendor today, the SOC 2 audit being still pending
How to use Archyl
A typical end-to-end flow, from setup to results.
- Create an account from the homepage, where no payment card is required
- Start a project with the '+' button in the sidebar or 'New Project' on the dashboard
- Name the project, which opens on an empty System Context diagram
- Add systems, then drill down to add containers, components and code elements
- Draw a relationship by dragging from one element's edge to another and describing the link
- Group elements with coloured overlays, which are specific to each C4 level
- Open Project Settings to connect a Git repository and run AI, ADR or documentation discovery
- Navigate by double-clicking to descend, using breadcrumbs to climb back and level tabs to jump
- Declare the architecture in an archyl.yaml file and version it alongside the code
- Wire the drift score into CI and connect the MCP server to Claude Code, Cursor or VS Code
Pros & Cons
Pros
- The C4 model is applied end to end rather than bolted onto a generic diagram editor
- Diagrams follow the code on every push, and the gap between the two is measurable
- The free plan is permanent and generous, with unlimited projects, docs, ADRs, contracts and MCP usage
- Security documentation is unusually thorough for a publisher this size: a dated whitepaper and a full DPA, both public
- EU data residency by default, in Belgium, with standard contractual clauses already published
- Bring-your-own AI provider removes every usage quota and leaves the AI contract with the customer
- Only code signatures reach the AI provider, never the full source
Cons
- SOC 2 Type 1 is described as an audit still pending, so there is no independent certification yet
- No postal address appears anywhere on the site and there is no legal notice page, only a company name, a legal form and a registration number
- A single person appears on the team page, which concentrates considerable key-person risk
- The trust centre promises EU processing while listing a US data centre on the same page
- The monthly and annual toggles show identical prices despite a 'Save 20%' badge
- Nothing published states whether customer data trains models, and no opt-out is documented
- Paid plans have no self-serve trial, and SSO and on-premise deployment are quote-only
Pricing & Plans
A permanent free plan is available: the Developer tier costs nothing, requires no payment card and is capped at 100 model objects and a single organisation. The cheapest paid tier is Business at 30 EUR per editor per month, followed by Scale at 45 EUR per editor per month, while the Custom tier is quoted on request. Payment is taken by card through Stripe, subscriptions may be cancelled at any time with access retained until the end of the billing period, and plan changes take effect immediately.
- 100 model objects
- 1 organisation
- 3 AI discoveries per month
- unlimited projects
- overlays
- docs
- ADRs
- API contracts
- everything in Developer plus 1
- 000 model objects
- 3 organisation invitations
- 10 AI discoveries and 200 AI chat requests per month
- share and embed links
- webhook events
- and Slack and Teams bots
- everything in Business plus unlimited model objects and organisation invitations
- 50 AI discoveries and 1
- 000 AI chat requests per month
- and 30 managed agent runs per month
- everything in Scale with all limits removed
- unlimited managed agent runs
- on-premise deployment
- priority support
- custom integrations
- SSO through SAML or OIDC
- and an SLA guarantee
Data, GDPR & hosting
A consolidated view of how Archyl handles your data.
GDPR overview
GDPR compliance is claimed explicitly and backed by documents rather than slogans. The trust centre carries a GDPR badge and the security whitepaper states full compliance. A Data Processing Agreement is published as a PDF for every customer, with standard contractual clauses, seven named sub-processors and thirty days' notice before any change to that list. Access, rectification, erasure, portability, objection and withdrawal of consent are all covered, with data export and cascade deletion built into the product. Breaches are notified within seventy-two hours under Article 33. Default storage is the EU, in Belgium, and is configurable. The publisher is established in France, so no Article 27 representative is required or named. The minimum age is sixteen, and privacy@archyl.com answers within five working days. SOC 2 Type 1 remains an audit in waiting, not a certification.
Who owns the data?
Under the terms of service you keep ownership of everything you create, upload or store: diagrams, documentation, decision records and API contracts. Archyl takes only a worldwide, non-exclusive, royalty-free licence to use, reproduce, modify and display that content, and strictly to deliver and improve the service. The platform itself, its features and its own content remain the publisher's property. In data-protection terms the roles reverse: the processing agreement makes the customer the controller and Archyl the processor, acting only on documented instructions. You can export the whole workspace and trigger a full cascade deletion of the account at any time.
Reuse rights
You may reuse anything you produce in Archyl without asking permission: ownership never moves, and a JSON export of the entire workspace is available on demand. On the publisher's side the privacy policy confines use to running the service, processing payments, sending technical and security notices, answering support, analysing usage, preventing fraud and personalising the experience. Personal information is explicitly never sold, and sharing is limited to six listed situations. AI features send only code signatures, meaning function names, type definitions and imports, to the AI provider rather than full source code, and the security whitepaper states that provider does not store them. Nothing published says whether customer data is ever used to train models.
Data retention & training
Hosting summary
Storage sits in the European Union by default, in Google Cloud's Belgian region (europe-west1), and that default is configurable through a server-side data-region setting. The trust centre nonetheless lists two data-centre locations, the EU in Belgium and the United States in Virginia, while stating on the same page that data is stored and processed in EU data centres; the processing agreement is the more precise document. File attachments live in private object-storage buckets encrypted with AES-256 and served only through short-lived signed URLs, with customer-managed keys possible, and self-hosted deployments may point at any S3-compatible provider and keep sole control of the bucket. Data is encrypted at rest with AES-256-GCM and in transit with TLS 1.3, and database connections require SSL. Seven sub-processors are named with their locations: Google Cloud Storage in the EU, and OpenAI, Stripe, Sentry, GitHub, GitLab and Bitbucket in the United States or split between the US and the EU. Transfers outside the EEA are covered by standard contractual clauses together with transfer impact assessments. The publisher is established in France.
Things to keep in mind
Risks and trade-offs to weigh before adopting Archyl.
- Generated documentation can look authoritative long before anyone has reviewed it: the AI proposes a model, it does not validate one
- A high drift score measures structural agreement with the code, not whether the architecture is any good
- Teams can stop debating architecture once a tool appears to hold the answer, quietly eroding the shared understanding the diagrams were meant to carry
- Exposing a whole architecture to AI agents concentrates sensitive design knowledge behind one set of credentials
- The claim that only code signatures reach the AI provider rests entirely on the publisher's own whitepaper, with no independent audit behind it
- Per-editor pricing quietly discourages the occasional contributor, who is often the person whose knowledge is missing from the model
- The publisher is a one-person French company with no independent certification yet, which is worth weighing before a critical system's documentation lives there alone
Setup & Integrations
Technical difficulty
Getting started is easy: an account, no payment card, and a first diagram within minutes. Real value arrives only once a Git repository is connected through OAuth, which usually needs someone holding rights on the organisation's GitHub, GitLab or Bitbucket. Beyond that the tool assumes a technical audience, treating C4, ADRs, DORA, MCP and a YAML DSL as known vocabulary. Writing archyl.yaml, wiring the drift score into a CI pipeline and configuring the MCP server are engineering tasks. On-premise deployment and SSO are Custom-plan projects rather than settings.
Deployment
Integrations
Supported languages
Behind Archyl
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement Archyl.
Frequently asked questions
What exactly does Archyl do?
Is there a genuinely free plan?
How does AI architecture discovery work?
What is the architecture drift score?
Where is my data stored?
Can AI coding agents use my architecture documentation?
How do I pay, and can I cancel?
Is there a minimum age?
Should you pick Archyl?
Archyl makes a clear bet: architecture documentation is worth doing only if it stays true, so the model is generated from the code and measured against it continuously. The bet is well executed. The C4 model structures everything rather than decorating a generic canvas, AI discovery removes the blank-page problem, and the drift score turns accuracy into a number a build can fail on. The agent tooling, with its MCP server, guardrails and scheduled runs, is more developed than the publisher's size would suggest, and the free tier is generous enough to be a real working plan rather than a demo.
The reservations concern the company more than the product. There is no postal address and no legal notice page, only a company name, a legal form and a French registration number. A single person appears on the team page. SOC 2 Type 1 is presented as an audit still pending, so the only certifications on display belong to sub-processors. The trust centre promises EU processing on the same page where it lists a US data centre, the monthly and annual price toggles show the same figures despite a discount badge, and the privacy policy and terms are more than a year older than the security documents beside them.
None of that is disqualifying, and the published DPA, the EU default region and the candour about what actually reaches the AI provider are better than many larger vendors manage. The honest summary is a well-designed, technically opinionated product from a very small French publisher: excellent for an engineering team that already thinks in C4, and worth a procurement conversation about certification and continuity before it becomes load-bearing.
- Choosing a selection results in a full page refresh.
- Opens in a new window.