Write for yourself.
Export for machines.

A Markdown editor that silently splits your writing into atomic blocks. Write freely. Each paragraph becomes a block. Every block is API-accessible.

Start writing → How it works
claim confidence: 0.92 · 3 links

Progressive disclosure eliminates the learning curve problem. Users write naturally; structure emerges from their words, not their effort.

supports: beginner-first-ux derived_from: obsidian-churn-data example_of: invisible-structure

How it works

01

Atomic blocks

Every paragraph becomes a typed block: note, definition, claim, evidence, question, summary. The system classifies automatically. You just write.

02

Semantic edges

Blocks connect through meaning: supports, contradicts, derived_from, example_of. Not just backlinks. Actual relationships that carry information.

03

AI memory

Three tiers: working memory for what's active, long-term for synthesized knowledge, reflection for self-improvement. Your graph gets smarter over time.

04

LLM-native export

One URL. Structured HTML with embedded JSON-LD. Hierarchical summaries. Any LLM with web access can read your knowledge at the depth it needs.

Knowledge that improves itself

Inspired by how human memory actually works. Not a filing cabinet. A living system.

Working Recently edited or referenced blocks. High detail, short-lived. What you're thinking about right now.
Long-term AI-generated summaries and abstractions. Distilled knowledge that persists across sessions and projects.
Reflection Insights about your knowledge itself. Repetition, gaps, weak assumptions, obsolete notes. The graph audits itself.

Zero copy/paste

Your knowledge, one URL away from any AI.

{
  "domain": "mark/product-research",
  "summary": "Competitive analysis of PKM tools...",
  "blocks": [
    { "type": "claim", "confidence": 0.92 },
    { "type": "evidence", "supports": "block_014" }
  ],
  "relations": [...],
  "trust_notes": ["verified_2026-05"]
}

Write what you know.

One idea per paragraph. Every paragraph a block. Every block reachable via API. No configuration, no AI inference, no friction.

Open the editor →