erode.dev

What It Is

Erode is an architecture drift detection tool for development teams. It compares code diffs against a declarative architecture model to identify structural changes such as undeclared dependencies, gateway bypasses, and new connections between services, surfacing them during local checks, CI, or code review. For example, if the frontend calls the User Service directly instead of going through the API Gateway, Erode will flag it as high severity and suggest either routing through the gateway or updating the model.

Core Capabilities

Its workflow is a multi-stage AI pipeline: it first parses the architecture components associated with the changed repository, then extracts dependency changes from the diff, compares them with the model, generates findings, and, when needed, produces model updates. The tool supports Gemini, OpenAI, and Anthropic: cheaper models are used for extraction and routing, while stronger models are used for analysis. It can run via GitHub Actions, CLI, Git hooks, and Claude Code Skill, and it also mentions experimental support for GitLab CI and Bitbucket Pipelines. In terms of model formats, LikeC4 is the recommended option, while Structurizr is experimental. It also supports monorepo scenarios where a single repository maps to multiple components.

Pricing and Open Source

The main content does not provide pricing plans for Erode itself. The page links to GitHub, an npm-based CLI package, and contribution documentation, but does not clearly state license information. In practice, users need to prepare a Gemini, OpenAI, or Anthropic API key, so the main cost comes from AI usage. The documentation also notes that a Claude Code hook may trigger checks on every edit, resulting in multiple API calls and potentially accumulating costs quickly.

Pros and Cons

Its main strength is its very specific positioning: it moves architecture review from “post-hoc archaeology” into the PR and local development stages. It can also turn LikeC4/Structurizr model changes into PRs, aligning well with the Architecture-as-Code approach. The documentation is relatively strong, with fairly complete commands, workflows, configuration, and example PRs. The downside is that adoption has significant prerequisites: teams need an existing usable architecture model and must accept the cost, latency, and availability constraints introduced by AI providers. Some integrations and formats are still marked experimental, and relationship deletions still require manual handling.

Who It’s For and Access from China

Erode is suitable for teams working with microservices, gateways, platform engineering, monorepos, or AI coding agents, especially in multi-person or multi-team environments where architecture boundaries can easily become difficult to control. The source text does not specify access conditions from China. Since it depends on external services such as OpenAI, Anthropic, and Gemini, teams in mainland China need to additionally evaluate network connectivity, enterprise compliance, API payment, and the availability of alternative models. If you only need diagramming, LikeC4, Structurizr, or traditional documentation tools may be enough; if you want to automatically detect architecture drift during code review, Erode is more purpose-built.