FSoft-AI4Code/CodeWiki

[ACL 2026] Open-source framework for holistic, structured repository-level documentation across multilingual codebases

What it solves

CodeWiki automates the creation of comprehensive, repository-level documentation for large codebases. It eliminates the manual effort of writing architecture overviews and module descriptions, especially for projects spanning thousands to millions of lines of code across various programming languages.

How it works

The tool analyzes a repository to build a dependency graph, which it then splits into a hierarchical tree of modules. AI agents are deployed recursively: leaf-module agents write individual pages based on the code, and parent-module agents synthesize these child pages into higher-level overviews. It also incorporates non-code artifacts like Dockerfiles and CI workflows into the graph to document build and deployment processes. The system supports incremental updates, allowing it to patch only the affected modules after code changes rather than regenerating the entire documentation set.

Who it’s for

Software engineers, architects, and maintainers of large-scale projects who need up-to-date, high-level documentation without the manual overhead of maintaining it.

Highlights

  • Multi-language support: Works with 11 languages including Python, Java, C++, Rust, and TypeScript.
  • Hierarchical decomposition: Scales to million-line repositories by clustering the dependency graph into a manageable module tree.
  • Artifact-aware: Documents build files, package manifests, and configuration schemas alongside the code.
  • Incremental updates: Reduces costs and time by only updating pages affected by specific code commits.
  • IDE Integration: Exposes an MCP server for use with AI IDEs like Cursor or Claude Desktop.
  • Multilingual output: Can generate documentation in various languages (e.g., Japanese, Vietnamese).

Related

  • Project
    langchain-ai/openwiki

    OpenWiki is a Node‑based tool that lets LLM coding agents automatically generate, query, and keep up‑to‑date a linked Markdown wiki inside a code repository. It integrates with many agents, supports parallel page generation, workspace linking, visual graph browsing, and CI‑driven updates, all while storing plain Markdown you own.

  • Project
    AIDotNet/OpenDeepWiki

    An AI-driven knowledge base system that converts Git repositories and local directories into structured documentation, searchable wikis, and MCP-enabled chat assistants.

  • Project
    divar-ir/ai-doc-gen

    An AI-powered documentation generator that uses a multi-agent architecture to analyze codebases and automatically create READMEs and AI assistant configuration files.

  • Project
    AsyncFuncAI/deepwiki-open

    DeepWiki-Open is an automated tool that analyzes code repositories to generate interactive wikis, visual diagrams, and guided codemaps for documentation.

  • Project
    chaitin/PandaWiki

    PandaWiki is an open‑source, Docker‑based wiki that adds large‑model AI for content creation, semantic search, and question‑answering. It supports rich editing, multiple import methods, and integration with chat tools, making it suitable for product docs, internal knowledge bases, FAQs, and AI‑assisted blogs.