# Add real-time chat to the dashboard A small, end-to-end feature: drop a chat panel into the existing dashboard so support engineers can talk to a customer while they're looking at the same data. The plan below walks through schema, transport, UI, and rollout. Approve the steps you want and tweak the rules — the agent will use your decisions before it starts. ## Architecture A quick look at how the pieces connect. The frontend opens a WebSocket through our existing edge gateway; the gateway routes to a stateless chat service that fans messages out via Redis pub/sub and persists to the existing events table. ```mermaid graph LR Browser["Browser"] -- WSS --> Edge["Edge Gateway"] Edge -- gRPC --> Chat["Chat Service"] Chat <--> Redis[("Redis pub/sub")] Chat -- INSERT --> DB[("events table")] Worker -- LISTEN --> DB ``` ## Backend ### Step 1: Add the `chat_messages` schema Create a new migration that adds: | Column | Type | Notes |--------------|-------------|---------------- | id | uuid | primary key | room_id | uuid | indexed | author_id | uuid | indexed | body | text | length-checked at write | created_at | timestamptz | default now() Add a partial index on (room_id, created_at DESC) WHERE deleted_at IS NULL to keep the recent-messages query fast. ### Step 2: Stand up the chat service A tiny stateless Go service. Subscribes to Redis, fans out to WebSocket connections. What's the right ceiling for messages per second per user? Bots will hit whatever we set; aim for something that feels natural to a human typing. - messages_per_second: 5 - burst: 20 - ban_on_violation: false ### Step 3: Wire it through the edge gateway The gateway already terminates TLS and authenticates users. Add a /chat route that upgrades to WebSocket and forwards to the chat service via gRPC, preserving the user's bearer token in a metadata header. (continues for ~1,800 more lines…)