Agents need to collaborate with humans. CRDTs give you real-time, offline-capable, conflict-free data structures. This post covers how to build a collaboration layer that both agents and humans can use.
Yjs for the data layer. It gives you:
Matrix for chat protocol. Matrix-CRDT bridges Yjs documents into Matrix rooms.
No great off-the-shelf CRDT kanban exists. Build it using:
TODO: Describe the architecture
┌─────────────────┐ ┌─────────────────┐
│ Human (browser)│ │ Agent (Temporal)│
└────────┬────────┘ └────────┬────────┘
│ │
▼ ▼
┌─────────────────────────────────────────┐
│ Yjs Document (CRDT) │
└─────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────┐
│ WebSocket Server / y-websocket │
└─────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────┐
│ Persistence (Postgres, IndexedDB) │
└─────────────────────────────────────────┘
TODO: How agents subscribe to CRDT changes
The agent needs tools that let it:
TODO: How to route notifications
When a CRDT change happens:
TODO: Step-by-step guide
TODO: Include working code examples