An AI documentation intern that works offline

What it does readmex ingests a project directory and emits a complete documentation starter kit: a README.md, a logo.png, a dependency manifest, a structure map, and script descriptions. It also compiles an interactive MkDocs site that can be served locally or deployed to GitHub Pages. The tool claims support for dozens of stacks—from Python and Rust to Nix and Vim—and can route LLM and image-generation calls to custom endpoints so you can keep everything on-premise.

The interesting bit The project bills itself as a Chinese-oriented open-source DeepWiki alternative, but the useful twist is seeing documentation generation as a batch pipeline rather than a chat session: one pass analyzes the code, generates assets, and builds a static site without further prompting.

Key highlights

• Claims coverage of 30+ language stacks, including Solidity, Nix, and Emacs Lisp.
• Can target local models by overriding LLM and text-to-image base URLs, though the README is fuzzy on exactly which local backends are supported.
• Generates auxiliary files beyond the README, such as a project-structure digest and script descriptions.
• Bundles a MkDocs wiki generator with flags for local serving and GitHub Pages deployment.
• Defaults to local embeddings for RAG-style context retrieval.

Caveats

• The README demands API keys for LLM and image generation even when running locally, so the “fully local” promise is more of a configuration aspiration than an out-of-the-box reality.
• Logo generation quality and multi-language support are explicitly flagged as unfinished on the roadmap.

Verdict Worth a look for maintainers who want a documentation scaffold without template archaeology, provided they are comfortable debugging local endpoint wiring. Skip it if you need hand-crafted architectural narratives or production-grade branding.