惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

Martin Fowler
Martin Fowler
有赞技术团队
有赞技术团队
博客园_首页
H
Help Net Security
GbyAI
GbyAI
aimingoo的专栏
aimingoo的专栏
V
Visual Studio Blog
The Cloudflare Blog
腾讯CDC
Jina AI
Jina AI
Last Week in AI
Last Week in AI
月光博客
月光博客
博客园 - 叶小钗
Google DeepMind News
Google DeepMind News
B
Blog RSS Feed
Blog — PlanetScale
Blog — PlanetScale
人人都是产品经理
人人都是产品经理
Engineering at Meta
Engineering at Meta
Y
Y Combinator Blog
Hugging Face - Blog
Hugging Face - Blog
博客园 - 聂微东
爱范儿
爱范儿
N
Netflix TechBlog - Medium
F
Fortinet All Blogs

InfoQ

GitHub Copilot Open-Source Project Brings Full iOS 27 Virtualization to Apple Silicon From Retrieval to Reasoning: Building Production-Ready Agentic AI Systems with Knowledge Graphs Lambda SnapStart Comes to Container Images, Ending a Packaging Tradeoff One Decade of Rustls: Evolution, Benchmarks, and Future Roadmap NVIDIA Personal AI Router Distributes AI Tasks across Local Compute Netflix Reworks Conductor for 420 Million Monthly Workflow Executions and 10X Larger Workflows tsgolint Reaches Stable v7, Bringing Go-Powered Type-Aware Linting to Oxlint Terraform AWS Provider Continues Rapid Expansion as AWS Infrastructure Becomes More Complex How To Run on Three Clouds at Once, and When Not To How LinkedIn Trains AI Job Search 8x Faster with Multi-Teacher Distillation Session Traces and Cost Controls Help Diagnose AI Agent Failures Advancing Embedded Go: Recoverable Panics, UEFI, Radio and Hardware Dev Kit Redefining GIS: Declarative Symbology and Collaborative Workflows in JupyterGIS Airbnb Cuts Authentication Code by 60% with Server Driven Architecture Kubernetes Promotes KYAML as a Safer, More Consistent Way to Work with Manifests Next-Gen Architecture Playbook: Insights and Patterns for the AI Era From S3 to GPU in One Copy: Rethinking Data Loading for ML Training Copilot Code Review Reaches Azure Repos, Billed Per Review with Reporting Two Days Behind Personality Over Skillset: How Adam Wachtel Builds Engineering Teams Tether: Apple Continuity Like Experience Between iOS and Linux Desktop Machines Twenty Years of jQuery: How a Little Library Rewired Web Development Shopify Introduces Gisting: Compressing LLM System Prompts into Learned Tokens Rigorous Yet Sustainable Human Reviews in the AI Era pnpm 12 Rewrites Package Manager in Rust, Accelerating Installs While Preserving pnpm 11 Workflows Instrumentation at Scale: Having Your Performance Cake and Eating It Too Cohere’s Parse 5 Promises Efficient Multi-Modal Information Extraction From Complex Documents Swiggy Uses 350+ Features and Multi-Task MLP to Predict Customer Lifetime Value OpenAI Details GPT-Live’s Architecture for Continuous Stateful Voice Interaction Beyond Prompting: Context Engineering for Production-Grade AI
Blume: Zero-Config Docs Framework That Turns a Markdown F...
Daniel Curtis · 2026-09-05 · via InfoQ

Blume, a zero-config documentation framework that renders a folder of Markdown into a complete docs site, has been released as open source by Hayden Bleasel, pitched as "a world-class docs framework for everything you ship" with zero-config setup, automatic SEO and AEO, 30+ components and no app boilerplate to write. Built on Astro and Vite and licensed MIT on GitHub, it has passed 1,300 stars and reached number seven on GitHub's TypeScript trending list.

The central claim is that the framework is the template, so there is nothing to clone. The CLI loads blume.config.ts, scans content into a graph and generates a hidden Astro project under .blume/ that it drives for dev and build. Node.js 22.12 or newer and one Markdown file are the only prerequisites, per the quickstart.

npx blume init
blume dev
blume build

Configuration is opt-in and type-safe, validated by a schema through defineConfig:

import { defineConfig } from "blume";
export default defineConfig({
 title: "Acme Docs",
 deployment: { site: "<https://docs.example.com>" },
});

Sites emit llms.txt and llms-full.txt, serve raw Markdown at any .md URL and can expose a hosted MCP server, while blume eval turns docs into a test suite: an agent answers user questions using only the documentation, a judge grades the answers and CI fails when the docs cannot answer.

On GitHub, TimonVS asked for route-based page tabs for design system docs, noting that using Starlight had forced "an ugly component override of the PageTitle component". Bleasel shipped query-parameter tabs within a day:

So the query-param version covers "shareable tab URL" and "multiple groups per page," but not "a separate indexed URL per tab."

Other issues reported dark mode elements flickering during client-side navigation and incorrect font fallback for Vietnamese text. Blume's own FAQ documents highlighted a formatting issue in Markdown with oxfmt collapses :::note directives onto a single line so they render as literal text, an upstream bug inherited from Prettier's Markdown printer, with a pinned patch as the workaround.

The changelog describes a bundled blume-migrate skill for Mintlify, Docusaurus, Fumadocs, Nextra and Starlight that translates source config into blume.config.ts, restructures content into filesystem-derived navigation with redirects for every moved route, rewrites callouts to ::: directives, converts icons to Lucide and points generated API references at openapi.sources. A third-party account exists too: PunGrumpy has written up moving Logixlysia from Fumadocs to Blume.

Against rivals, Blume's comparison table frames Mintlify as a hosted platform with a closed core, and Fumadocs, Nextra and Docusaurus as a library plus an app you scaffold and then maintain on a React runtime. Blume claims the middle ground: open source, hostable anywhere, with only Markdown to own and a core theme that ships zero client framework JavaScript. Escape hatches include component overrides and blume eject, which promotes the generated runtime into a standalone Astro app.

Blume ranked second of the day on Product Hunt with 304 upvotes shortly after launch, and MarkTechPost covered it in July. On the Product Hunt launch, someone asked:

curious what "AI-ready" means concretely here - is it generating something like an llms.txt under the hood, or is it more about the markdown being cleanly chunked/structured for retrieval.

To which Bleasel replied:

Means lots of things - llms.txt, raw markdown via negotiation headers, mcp servers, agent readability, skills. All in the pursuit of making your docs easier to read by agents. Check it out here: https://useblume.dev/docs/configuration/ai

Blume is an open-source documentation framework distributed under the MIT license and maintained by Hayden Bleasel, an Australian design engineer whose other projects include Ultracite and Files SDK. It renders static HTML from a folder of Markdown on Astro and Vite, ships a core theme with no client framework JavaScript, and is available on npm, with documentation at useblume.dev.

About the Author

Daniel Curtis