Architecture

How the editor and document engine packages fit together.

Lextrix splits into an editor surface and a document engine. The editor is how users type. The engine is how you store history, sync clients, and review AI edits.

Plain text

ChangeSet → Document / Handle / Version
                 ↓
            Editor (DOM)

lextrix-change is the engine core. It does not depend on Postgres, WebSockets, or your auth system. Collaboration protocol lives in lextrix-collab. The reference host that wires Postgres + WebSocket is lextrix-server. AI sits in lextrix-intelligence and only produces proposals.

A few rules that show up everywhere:

A ChangeSet is a transition; a Document is a value; a Version is history. Editor undo is none of those.
Concurrent edits use OT. The server picks Version ids and accept order.
Durable writes go through compare-and-append (changeId, owner_epoch).
Snapshots are checkpoints, not a second HEAD.
Presence is ephemeral — it does not become document history.

Deeper ADR material stays on GitHub. Next: Document Model.

Lextrix splits into an editor surface and a document engine. The editor is how users type. The engine is how you store history, sync clients, and review AI edits.

Plain text

ChangeSet → Document / Handle / Version
                 ↓
            Editor (DOM)

A few rules that show up everywhere:

A ChangeSet is a transition; a Document is a value; a Version is history. Editor undo is none of those.
Concurrent edits use OT. The server picks Version ids and accept order.
Durable writes go through compare-and-append (changeId, owner_epoch).
Snapshots are checkpoints, not a second HEAD.
Presence is ephemeral — it does not become document history.

Deeper ADR material stays on GitHub. Next: Document Model.