snowflake-architecture-diagram-connected-ui

Connected-UI mode

Defers to the live SnowGram Next.js frontend at localhost:3002 (backed by the FastAPI service at localhost:8082) for the full ReactFlow + ELK rendering experience and multi-tab conversational refinement.

This mode does NOT spawn the lightweight viewer. The user gets pixel-identical SnowGram output, but at the cost of two running processes.

Prerequisites

• Connected-CLI prerequisites (connection + agent reachable)
• SnowGram backend running at http://localhost:8082
• SnowGram frontend running at http://localhost:3002
• frontend/.env.local populated with a valid SNOWFLAKE_PAT for the active account
• backend/.env populated with connection details

Workflow

Step 1: Probe both servers

curl -s -o /dev/null -w "%{http_code}" --max-time 1 http://localhost:8082/health || true
curl -s -o /dev/null -w "%{http_code}" --max-time 1 http://localhost:3002/ || true

Backend	Frontend	Action
200	200	Continue to Step 2
200	down	Print frontend start command, stop
down	200	Print backend start command, stop
down	down	Print both start commands, stop

Do NOT auto-start servers. They may need credentials, port adjustments, or other manual setup.

Start commands to print:

# Backend (terminal 1)
cd <repo>/backend
source venv/bin/activate
python -m api.main

# Frontend (terminal 2)
cd <repo>/frontend
npm run dev

⚠️ STOP: tell the user which servers are down and what to run, then exit. They will resume the skill once both are up.

Step 2: Optional pre-seed via agent

If the user provided a prompt, invoke the agent once via CLI to pre-seed the canvas before opening the browser:

cortex agents run SNOWGRAM_DB.AGENTS.SNOWGRAM_AGENT "<user's prompt>"

Capture the response and POST it to the running backend so the frontend has a default tab populated:

curl -X POST http://localhost:8082/api/diagram/seed \
  -H "Content-Type: application/json" \
  -d '{"mermaid": "<extracted>", "title": "<short>"}'