Skip to content
ArchBeagle
Architecture intelligence for agentic development

Agents write code.
Keep your architecture in control.

Understand your codebase. Review architectural changes.
Track agent work alongside the checks that verify it.

Code context · Architecture PR review · Agent visibility

Explore the context behind every change.
Example: acme / shopIllustration · sample data

See how your code connects

Select a component to explore its context.

─ Indexed relationship
Click any node to inspect its context.

The examples on this page are illustrations with sample data. They show what ArchBeagle tells you, not its screens.

Function-level intelligence across your stack
PythonTypeScriptJavaScriptGoJavaPHPRuby
Architecture-aware pull request review

A small diff can make
a big architectural change.

Review the code in the context of components, dependencies, contracts, and the decisions your team approved.

◇ Architecture reviewIllustration · sample data

Simplify checkout calculation #142

acme/shop · claude/checkout → main · 3 changed files · model v12

! Architecture violation
Changed files → components
src/shop/web/checkout.pyComponent: Web
  16  # Calculate the checkout total− 17  from shop.orders.service import get_total+ 18  from shop.data.orders import OrderRepository  19+ 20  total = OrderRepository().cart_total(cart_id)

The new import connects Web directly to Data. The approved model routes that dependency through the Orders service.

Evidence from a check. Context from your model. A clear next action.

Trace every finding

Go from a review result to the rule, component, file, and decision behind it.

Distinguish drift from violations

A changed model relationship and a forbidden dependency deserve different responses.

Keep the review connected

Read agent reports alongside independent check evidence on the linked pull request.

Agent activity & verification

What your agents reported.
What your checks confirmed.

Follow connected agent sessions across people, repositories, and pull requests. Keep reported progress and verified outcomes side by side.

◎ Agent work / acme engineeringIllustration · sample data
3Reported sessions
2Finished sessions
1Architecture check failed
1Architecture check passed

Identity comes from the reporting token. Session summaries, tests, and artifacts are agent-reported. Check evidence is shown separately.

Visibility covers connected agents and reported sessions. It does not imply all agent activity is captured.How reporting works →

Accountable by identity

Associate each reporting token with an agent, a person, and optionally a team.

A history you can inspect

Follow started, progress, finished, and abandoned sessions with links to their work.

Claims stay labelled

Keep test claims, verification notes, and artifacts distinct from repository checks.

Session evidence

Illustration with sample data: a session report, and separate repository-check evidence.

Built for your coding agents

Better context.
Before the first edit.

Your agent should know what a function touches and which boundaries matter. ArchBeagle gives it the same code intelligence your team sees, through MCP.

  • Find symbols, trace calls, and inspect change impact
  • Check planned imports against approved architecture
  • See index freshness and resolution limits
Connect your agent →

Claude Code · Codex · Cursor · One MCP server

claude-code → archbeagle / MCPIllustration · sample data
› I'm changing discount rounding.
What depends on Cart.total?
change_impact({
  repo: "acme/shop",
  symbols: ["shop.orders.cart.Cart.total"]
})
→ 6 affected symbols. 3 related tests.

Checkout and invoice rendering call this function.
Review their rounding expectations before editing.

Decisions, components & rules

Make the reasoning
part of the review.

Connect architectural decisions to the components they govern, the dependencies they allow, and the findings that need attention.

ADR-012 · AcceptedOwner: Commerce

Business rules belong in the service layer.

Checkout should call the Orders service. Direct access to the data layer would bypass the place where order validation and totals are coordinated.

Web → Orders → Data
Related deterministic check
ARCH-DEP-3 · Undeclared component dependency

The check compares imports with the dependencies declared in the pinned model. The linked ADR explains why the boundary exists.

See this decision in a PR review →

Illustration · sample data Decisions provide context; enforcement depends on the corresponding machine-checkable rule.

API, data, tool & event contracts

Catch a changed interface
before it surprises a consumer.

A removed endpoint, a required input, a deleted field. Make interface changes visible in the same review as the code.

⇄ Contract change reviewIllustration · sample data
Before · mainopenapi.yaml
GET /orders/{id}
  parameters:
    - name: include_items
      in: query
      required: false
After · pull requestBreaking change
GET /orders/{id}
  parameters:
    - name: include_items
      in: query
      required: true

CON-BRK-1 · An optional query input becomes required.

Clients that omit include_items no longer satisfy the contract. Keep the input optional or plan a versioned migration.

Contract findings warn on the pull request. They do not fail the check.Rule catalogue →
Architecture drift & finding history

A finding should have
a history, not just a count.

Keep the same finding identity across runs. See when it appeared, when a default-branch check cleared it, and whether it returned.

  • Distinguish rule violations from model drift
  • Follow open, fixed, and reopened findings
  • Spot stale checks, missing owners, and ageing gaps
Explore the finding lifecycle →
Finding historyIllustration · sample data
Finding AB-1042 · OpenFirst seen in PR #142. Still reported on the checked revision. A passing check on a corrected PR is useful evidence; resolution is confirmed when a default-branch run no longer reports the issue.
Connect the tools you already use

One shared context.
Across your workflow.

Bring in repositories, architecture models, and documentation. Connect coding agents and send findings where your team works.

GitHubAvailable

A GitHub App: repositories, pull requests, reviews, labels, CODEOWNERS and the "architecture" check run.

Code hosting and CI

GitHub ActionsAvailable

The check runs in your workflow, and deploy gates verify GitHub Actions OIDC tokens.

Code hosting and CI

GitLabComing soon

Merge requests, pipelines and sign-in, after GitHub.

Code hosting and CI

Claude CodeAvailable

The MCP server and a blocking hook before each edit.

AI coding agents

CodexAvailable

The MCP server and a blocking hook before each edit.

AI coding agents

CursorAvailable

The MCP server and a blocking hook before each edit.

AI coding agents

GitHub CopilotComing soon

The MCP server, the agent kit and its instruction files.

AI coding agents

GitHubAvailable

Sign-in for everyone, personal accounts included.

Single sign-on

Okta, Microsoft Entra ID, GoogleAvailable

Any OpenID Connect provider, with allowed email domains.

Single sign-on

SlackAvailable

A daily digest and an alert for each new critical finding.

Alerts and work tracking

GitHub IssuesAvailable

Tasks and critical findings as issues, closed when they are resolved.

Alerts and work tracking

Microsoft TeamsComing soon

The same digest and alerts as Slack.

Alerts and work tracking

JiraComing soon

Tasks and findings as Jira issues, opened and closed with them.

Alerts and work tracking

StructurizrAvailable

Workspace JSON, read as your model.

Architecture models

C4-PlantUMLAvailable

C4 diagrams in PlantUML, read as your model.

Architecture models

LikeC4Available

A LikeC4 model, with its views and layers.

Architecture models

BackstageAvailable

Read the software catalog, and export the model back to it.

Architecture models

OpenAPI and AsyncAPIAvailable

API contracts compared on every pull request, with breaking changes flagged.

Contracts and supply chain

CycloneDX and SPDXAvailable

SBOMs read for dependency and licence policies.

Contracts and supply chain

ConfluencePro

Pages read as documentation sources.

Documentation

NotionPro

Pages read as documentation sources.

Documentation

Google DrivePro

Docs read as documentation sources.

Documentation

Argo CD and FluxPro

A deploy gate before each sync, and the deployed state.

Deployments and runtime

OpenTelemetryPro

Runtime calls over OTLP, from any collector or backend, compared with the model.

Deployments and runtime
Your code stays yours

Keep the context.
Not the source code.

Read the security details →
◇

Isolated indexing

Repositories are read in a sandbox. ArchBeagle stores paths, names, line ranges, and metrics—not your source code.

◇

Access with clear boundaries

Reads are scoped to your organisations. Each agent token belongs to an agent and person, and is stored as a hash.

Fits the way you already work

Start with a repository.
Build a shared understanding.

Keep GitHub, your editor, and your agents. Add context that connects them.

01

Connect the repositories you choose

Sign in with GitHub and install the ArchBeagle App for your account or organisation.

02

Explore your code and architecture

Inspect the code index. Bring an architecture model from Git, or approve a draft from the code.

03

Guide agents and check changes

Connect MCP for agent context and session reporting. Set up architecture checks in your CI workflow.

Browse integrations by workflow →
A few things worth knowing

Good questions.
Clear answers.

Need more detail? Explore the docs →

How does this fit alongside a code reviewer?

ArchBeagle focuses on code relationships and your approved architecture, before edits and during pull requests. Use that context alongside the bug detection, security checks, and review conversations already in your workflow.

Do I need an architecture model to get started?

No model is needed for the code index. Connect your chosen GitHub repositories to explore symbols, relationships, and complexity. Architecture enforcement needs a model: one maintained in Git, or a draft generated from your code that you review and approve.

Which languages and agents are supported?

Function-level indexing covers Python, JavaScript, TypeScript, Go, Java, PHP, and Ruby. The architecture check supports Python repositories today. Claude Code, Codex, and Cursor connect through MCP. Index coverage, architecture checks, and hook support are separate capabilities; call resolution also varies by language.

Does the agent dashboard verify everything an agent says?

No. Sessions, test results, verification notes, and artifacts reported by an agent are labelled self-reported. Repository check results are separate evidence. A finished session is not automatically a passed check, and the dashboard only shows connected and reported activity.

Where does my source code go?

Indexing reads your repository’s main branch in a sandbox and stores paths, names, line ranges, and metrics, rather than source code. Architecture checks run in your CI; ArchBeagle reads the resulting build artifact through its GitHub App.

How are architecture checks enforced?

The architecture check in your CI fails on violations; drift alone does not fail it. Require that CI job in GitHub to use its result for merge protection. Contract findings warn on the pull request and do not fail it.

Build with the bigger picture

Give your next change
the context it deserves.

Bring code context, architecture review, and agent evidence into one shared view.

Get started with GitHub →Choose your repositories. Explore your code. Set your boundaries.