llmchat

AI-first support widget + admin inbox. Standalone product built on LLM Gateway.

Stack

• Runtime: workerd via Ploy
• Database: D1-compatible SQLite (Ploy db: binding) with Drizzle ORM
• Cache / rate-limit: Ploy state: binding (KV-compatible)
• Inference: LLM Gateway (single backend, configured via LLMGATEWAY_API_KEY)
• Email: Resend REST API (outbound) + inbound webhook
• Auth: Better Auth (email + password)
• Billing: Stripe (metered AI messages + seat-based)

Layout

apps/
  api/          # Hono — public chat (SSE), dashboard API, webhooks, serves widget.js
                #   migrations/ — D1 SQL migrations (Ploy applies these on dev/deploy)
  dashboard/    # Next.js — admin inbox + project settings
  marketing/    # Next.js — llmchat.io landing
packages/
  db/           # Drizzle schema + client; emits SQL into apps/api/migrations
  shared/       # Zod schemas + shared types
  widget/       # Vite lib build → widget.js served by api

Develop

pnpm install
cp apps/api/.env.example apps/api/.env   # fill in keys before running
pnpm dev                                 # = ploy dev — boots all apps

ploy dev at the repo root runs all projects together: api on :8787, dashboard on :3001, marketing on :3002, and the shared Ploy dashboard on :9787.

Migrations live at apps/api/migrations/ (Ploy convention) and are applied automatically by ploy dev. To regenerate from the Drizzle schema:

pnpm migrations   # drizzle-kit generate (writes new SQL into apps/api/migrations)