AxiomCode Graph maps relationships in a repository so developers and coding agents can ask what a change reaches, rather than relying only on text search. Its 0.2.0 release, published October 14, adds an HTML/CSS graph: pages, loaded stylesheets, imports, inline styles, and CSS precedence rules.

  • Concrete use: Ask which pages or elements a class, ID, CSS variable, or stylesheet change could affect. The new graph models specificity, cascade layers, !important, and custom properties across files. Java, TypeScript, and Python already have code graphs; JavaScript, C#, HTML, and CSS are marked beta.
  • Setup and interface: Install npm i -g @axiomcode/code-graph@0.2.0; the documented requirements are Node 22.13 or newer and Python 3. Use axiomcode index, axiomcode impact <name>, or axiomcode path <A> <B>. Agents can connect through Model Context Protocol (MCP), a standard tool interface, or supported agent plugins. Add .axiomcode/ to .gitignore.
  • Inspectable output: The graph lives in a local SQLite database. Code-call relationships distinguish resolved targets, multiple possible targets, library boundaries, and unresolved calls. A graph refresh can leave queries reading the previous version; --fresh waits for the rebuild.
  • Access and cost: The release includes downloadable packages and prebuilt engines for Mac, Linux, and Windows. The reviewed documentation gives a local installation path, not a hosted per-query price. The license permits internal use but restricts competing commercial products/services; it becomes Apache 2.0 for each version after two years. This is source-available, not unrestricted open source.

Important boundary: The new web graph links HTML and CSS, but does not resolve scripts and event handlers into JavaScript call chains. It should not be treated as a complete model of a dynamically rendered application. Likewise, a checked graph edge is not proof that an agent’s proposed edit is correct. The project’s benchmark results are maintainer-reported; this capture reviews the release and documentation, not a hands-on test.