.ember
A document format where files are their own database, their own transaction history, their own routing layer. The engine behind every Binary Soul product.
Not markdown. Not JSON.
Ember is structured text optimized for how AI actually processes information.
Self-describing
Every .ember file carries its own metadata, relationships, and semantic routing in a single anja header line. No external schema needed.
// anja: overview::page // ~active <-"project arch" // @03-06 #b1
Graph-native
Typed edges connect documents: extends, settles, leads, blocks. The file system becomes a queryable knowledge graph.
>settles(old-decision) >extends(parent-plan) >leads(next-step)
88x denser
The same content that takes 5,290 tokens in JSON takes 60 in .ember. AI gets 88x more room to think about what matters.
@page "Overview" id=b1 h2| Architecture p| The core system... li| First component
Where .ember lives
Pro tier translates JSON responses into .ember compact text. Data arrives pre-structured.
3,705 .ember files. 8,500+ anja edges. Zero traditional database. The format IS the backend.
From the format itself to the computation it directs. Filed March 30, 2026.
One binary. Everything.
ember-core compiles to a single Rust binary called ember-serve. It reads .ember files off disk and serves a complete production application — no database, no framework, no cloud dependencies.
This site, the Grimoire, and the API all run on the same binary on the same machine. Not a demo. Not a pitch. The thing you're using right now.