docmd

What docmd actually is

docmd (github.com/docmd-io/docmd) is a Node CLI that turns a folder of Markdown into a documentation site. `npx @docmd/core dev` runs in any folder with Markdown in it and serves on localhost with no config file, no frontmatter requirement and navigation derived from your directory structure; `npx @docmd/core build` writes a static site you can host anywhere. The licence is a plain MIT in the repository's LICENSE file, the project started in May 2025, and the release history is steady rather than bursty — 0.8.15 on 15 July 2026 through 0.9.4 on 23 August 2026, with 14 open issues at capture.

• It starts with nothing configured No config file is required at all. A `docmd.config.json` exists for when you want to set a title, a canonical URL, or a different source and output directory, and it can be written in TypeScript or JavaScript when you need dynamic values.
• It emits AI-facing outputs as core plugins, not add-ons The `llms`, `okf`, `search`, `seo` and `sitemap` plugins ship enabled by default, so the context files and the sitemap are produced by an ordinary build rather than by a step someone has to remember.
• It has a migration path in the box `docmd migrate` converts an existing Docusaurus, VitePress, MkDocs or Starlight setup, which matters mostly because it means trying it does not start with a rewrite.
• It handles locales and versions as first-class builds Multi-locale builds produce a per-locale search index, per-locale context files and hreflang tags, and doc versions are native rather than a plugin. If you are already translating a site, this is the part that decides whether it fits.
• Check the Node version against the package, not the README The README states Node.js 18 or newer; the published `@docmd/core` package declares `engines.node >= 20.0.0`. On a machine still running 18 the install is what will tell you, so plan for 20.

There is a hosted side too, and it is worth knowing where the line falls before you adopt it. The AI Assistant plugin does retrieval-augmented chat over your own docs using your own API key or a local model; the Cloud Relay at cloud.docmd.io exists so a purely static site can run that assistant without you standing up a backend. The compiler and every artifact below are in the MIT repository — the relay is the optional hosted convenience, which is the open core shape in miniature.

The four machine-facing artifacts, and what reads each

Here is why that list is worth more than it looks. We publish `llms.txt` and `llms-full.txt` for AgentCeres — the AI Growth Officer, at agentceres.com — and we wrote both by hand. They drifted. The files listed internal roles a customer cannot actually select, and a per-seat pricing table describing a model we have never sold. Nothing caught it for weeks, because the checks that guard our marketing copy scan the application's source directories, and those two files live in a static `public/` folder that none of them read. A generated context file removes that whole class of mistake by construction — not because generation is cleverer, but because there is no second copy left to fall out of date.

Where it fits when you are trying to get found

Documentation is one of the few things a small team writes that an answer engine can use directly: it is specific, it is dated, and it describes a product nobody else can describe as accurately. That makes it good raw material for Generative Engine Optimization, and docmd lowers the cost of making it retrievable to roughly zero. If your question is the broader one of how to get your startup cited by ChatGPT, this is one input to that, not the answer to it.

Be clear about the limit, because it is easy to overrate. Emitting a context file makes your docs cheap to retrieve; it does not make a model choose you, and no build step can. It also only covers the docs. Your pricing page, your comparison pages and your landing pages are a different corpus with different problems, and they are usually the ones a buyer's question actually lands on. If the job you have is a documentation site rather than the machine-readable half, Docusaurus is the more conventional pick and is covered separately.

The honest sequencing for a founder: build the docs because your users need them, take the context files as a free by-product, and treat the traffic as a second-order effect that may take months to show up. Publishing the files is a build flag. Being worth citing is the work, and that part is still writing — which is where AgentCeres drafts alongside you and a human approves anything that goes out.