Markdown · for AI agents · weighs nothing

Sites that hover.

A markdown CMS your AI agent can edit directly. Articles are plain .md on disk — Claude, Cursor, any LLM with file access reads and writes them like a human. No DB on the hot path. No build step. No plugin theatre.

Agent-friendly markdown AEO schemas (13+ types) MIT licensed
/ RENDER
3.2ms
P50 render target (benchmark suite ships v3.1). Markdown parsed once, cached in memory. The hot path never touches a database.
/ 01
/ SOURCE

Article meta as YAML.

articles/best-ai-tools.meta.yaml
---
title: "Best AI Tools 2026"
primary: "best ai tools"
secondary:
  - ai productivity
  - ai comparison
hero_image: /assets/ai.webp
# pillar-cluster auto-wires related
---
/ 02
/ ANTI-BOT
nginx/SQLite
Log parse Auto-blacklist Sandbox UI Zero plugins
/ 03
/ WEIGHT

Featherweight stack.

Tight engine — small enough to read end-to-end in an afternoon. No JS framework. No build step. The whole engine fits in your hand luggage.

WordPress
~85 MB
Ghost
~55 MB
Bird CMS
~2 MB
/ 04
/ ENGINE

Origami runtime.

Every site is a folded copy of the same engine. Symlink-swapped releases, checksum-gated rollback.

/ 05
/ DEPLOY

Lands without a bump.

$ update.sh 3.0.0
→ smoke-test: PASS
symlink swapped, live
/ 06
/ AEO

Built for AI citation.

FAQ, HowTo, Review, Comparison, Article — the Schema.org types AI engines actually cite. Plus auto-generated llms.txt and pillar-cluster interlinking for English SEO-content patterns.

Article FAQPage HowTo Review Comparison BreadcrumbList LocalBusiness Service
/ 07
15ms
Hummingbird wingbeat
vs.
3.2ms
Bird CMS render
5 renders per wingbeat.
The site is faster than the bird.
/ MANIFESTO / 2026
A site shouldn't take off — it should already be in flight. Bird is built on the assumption that the next decade of writing is half-human, half-agent, and both should reach for the same plain text. So we cut the weight: no DB on the hot path, no framework, no plugin theatre. Just markdown, on disk, ready in a wingbeat.