Introducing Blume: an open supply, zero-configuration doc framework that sends AI-enabled paperwork from Markdown folders.

Hayden BreazellLaunched by OpenAI professional builders Blumean open supply documentation framework. Blume was shipped to npm on the identical day as model 1.0.3. It is as simple as dropping Markdown right into a folder and distributing your documentation web site. App boilerplates should not created or maintained after the actual fact. This undertaking is MIT licensed and open supply.

What’s Blume?

Blume is a command line instrument mixed with a part library for documentation. Learn a folder of Markdown or MDX recordsdata. A production-grade documentation web site might be generated from that folder. Its output contains navigation, search, theming, and Open Graph photos. Configuration stays elective and is added one file at a time. The code is a TypeScript monorepository. The printed package deal is situated at packages/blume. Blume’s personal documentation is: apps/docsconstructed with Blume itself. Requires Node.js 22.12 or later. Runs on Bun, pnpm, npm, or yarn.

How Blume works?

Beneath the floor, Blume generates and promotes a hidden Astro undertaking. First, the CLI is loaded blume.config.ts Scan and graph your content material. Subsequent, create the Astro undertaking .blume/ listing. Astro renders all pages by a single catchall route. This route imports Blume’s shipped elements, generated knowledge, and overrides. Each time you run .blume/ might be regenerated and solely modified recordsdata might be overwritten. In consequence, sizzling reloading whereas enhancing stays quick. The core theme doesn’t include shopper framework JavaScript. In consequence, pages rating excessive on Core Internet Vitals by default. If you’d like full management, blume eject Promote your runtime to a standalone Astro app. Eradicated initiatives nonetheless rely upon blume package deal.

Begin

Setup is accomplished with a single command, so onboarding is fast.

after that, blume dev Begin the new reload server. in the meantime, blume construct Writes static HTML and native search index dist/. The configuration file is precise TypeScript and is schema-validated.

// blume.config.ts
import { defineConfig } from "blume";

export default defineConfig({
  content material: {
    sources: [
      { type: "filesystem", root: "docs" },
      { type: "notion", database: process.env.NOTION_DB },
    ],
  },
});

The configuration is entered so the editor will catch the error earlier than the construct runs. CLI covers your complete lifecycle past these fundamentals.

AI-enabled design

Blume targets brokers in addition to human readers. Including all pages returns uncooked markdown .md to that URL. a single flag fires llms.txt and llms-full.txt For brokers. Every web page may be copied as Markdown or opened in ChatGPT, Claude, or v0. An elective in-page Ask AI assistant solutions readers’ questions straight. Runs on high of the AI ​​SDK by way of Vercel AI Gateway, OpenRouter, Inkeep, or any OpenAI appropriate endpoint. Blume can even host Mannequin Context Protocol (MCP) servers. By way of this, Claude Code, Cursor, and VS Code learn the doc straight.

claude mcp add --transport http your-docs https://docs.instance.com/mcp

This server exposes 4 read-only instruments. search_docs, get_page, list_pagesand get_navigation.

Utilization and examples

These capabilities are mapped to concrete jobs. For API merchandise, drop an OpenAPI or AsyncAPI specification. Blume then renders an interactive reference containing the schema, authentication, and request playground by way of Scalar. For libraries, level Blume to the GitHub launch. Every launch is summarized in a changelog timeline generated with an RSS feed. If you wish to attain a worldwide viewers, add translated recordsdata for every locale. Blume helps 36 locales, locale-aware routing, and right-to-left format. For blended content material, mix native recordsdata with distant MDX, Notion, or Sanity. All sources are rendered by the identical part.

evaluate Blume

For context, listed below are Blume for 3 widespread documentation approaches. Options change shortly, so please assessment present particulars earlier than hiring.

Benefits and drawbacks

Strengths

• Begin with zero configuration: A Markdown folder turns into a whole web site.
• Static-first output doesn’t include core shopper JS and helps Core Internet Vitals.
• Constructed-in AI surfaces: llms.txtper-page markdown, MCP Server, and Ask AI.
• It is a type-safe configuration, so the editor flags errors earlier than working the construct.
• Eject path to standalone Astro app reduces long-term lock-in.

Weak point

• Model 1.0.3 is new, so the ecosystem continues to be younger.
• Requires Node.js 22.12 or later, which can be lacking in some environments.
• Request-time options resembling Ask AI and MCP require a server adapter.
• A self-hosted mannequin means you personal the analytics and assistant wiring.
• It has fewer third-party integrations than at the moment’s mature hosted platforms.

Vital factors

• Blume turns your Markdown folder right into a manufacturing documentation web site with no configuration.
• It drives hidden Astro and Vite initiatives and may be ejected into standalone Astro.
• Constructed-in AI capabilities. llms.txtper-page markdown, and MCP server.
• It’s MIT licensed, requires Node.js 22.12 or larger, and reached npm v1.0.3 on launch day.
• Early adopters embrace Quiver, which migrated from Mintlify, and Neon’s add-mcp documentation.

Please test GitHub repository. Please be happy to observe us too Twitter Do not forget to hitch us 150k+ML subreddit and subscribe our newsletter. grasp on! Are you on telegram? You can now also participate by telegram.

Have to associate with us to advertise your GitHub repository, Hug Face Web page, product launch, webinar, and many others.? connect with us