https://kidplayai.techseeding.com.au/
An AI-powered craft maker platform for kids aged 8–12 who love games, science, engineering, and AI. Kids describe the craft they imagine, then watch a real AI agent think, design, and build it step by step. Rather than hiding AI behind a polished UI, KidPlayAI surfaces the raw interaction so kids see exactly how AI reasons, creates, and solves problems.
- A kid visits the homepage and clicks Start Making Crafts
- They enter their name on the login page and request access
- An admin approves the request from
/admin - The kid is dropped into a sandbox: a live craft preview on the left, a streaming chat with the AI on the right
- They type natural language requests (e.g. "make a craft where I catch falling stars") and watch the AI think, call tools, and edit a single
index.htmlin their sandbox - The preview iframe refreshes each time the AI writes the file
- Frontend: React 19, Ant Design 6, Vite, react-router-dom v7, react-markdown
- Backend: Node.js, Fastify, WebSockets
- AI Agent: in-process agent loop via the Vercel AI SDK (
ai+@ai-sdk/deepseek), backed by DeepSeek. One Fastify process serves every chat session — no per-kid subprocess - Database: PostgreSQL with Drizzle ORM
- Infrastructure: AWS (ECS Fargate, Aurora Serverless v2, EFS, ALB) provisioned with CDK v2
- Package Manager: pnpm workspace monorepo
src/api/ Fastify backend (routes, db schema, sandbox manager, ws)
src/portal/ React frontend (Vite)
deploy/ AWS CDK app
devops/ Production Dockerfile + entrypoint
docs/ Schema and design references
- Node.js (with
node --env-filesupport) - pnpm 10+
- A local PostgreSQL with a
kidplayaidatabase - A DeepSeek API key (for the AI agent inside each sandbox)
pnpm install
cp .env.sample .env # then fill in KPAI_* values
pnpm db:migrate # apply schema migrations
pnpm start # API on :9511, portal on :9512Open http://localhost:9512.
pnpm start # run API (:9511) and Vite portal (:9512) concurrently — Vite proxies API/WS calls to the API server
pnpm start:api # just the Fastify API with --watch and --inspect, loading .env
pnpm start:portal # just the Vite dev server (portal workspace)
pnpm db:generate # generate a Drizzle migration from schema changes in src/api/db/schema.js
pnpm db:migrate # apply pending migrations to the local database
pnpm db:studio # open the Drizzle Studio GUI to inspect/edit local datapnpm build:prod # bundle the portal into dist/public/ and copy the api into dist/src/
pnpm build:docker # run build:prod, then build the techseeding/kidplayai Docker image
pnpm start:prod # run the production server from dist/, loading .env.production
pnpm start:local:docker # build and run the production image locally on :9515 with .env.localdocker
pnpm release # build the image and deploy via the CDK app under AWS_PROFILE=kpai
pnpm db:connect:prod # open a psql session against the production Aurora database
pnpm db:jdbc:prod # print a JDBC connection string for the production databaseAll env vars are prefixed with KPAI_. See .env.sample for the full template.
| Variable | Description |
|---|---|
KPAI_DATABASE_URL |
PostgreSQL connection string |
KPAI_API_PORT |
Port the API server binds to |
KPAI_PUBLIC_URL |
Public-facing app origin |
KPAI_JWT_SECRET |
Secret for signing JWTs |
KPAI_SANDBOX_DEEPSEEK_API_KEY |
DeepSeek API key used by the in-process sandbox agent |
KPAI_SANDBOX_DEEPSEEK_MODEL |
DeepSeek model id (optional, defaults to deepseek-chat) |
pnpm start loads .env; pnpm start:prod loads .env.production.
Production runs at kidplayai.techseeding.com.au on AWS ap-southeast-2. Infrastructure is defined in deploy/ using AWS CDK; pushes to main deploy via the GitHub Actions workflow in .github/workflows/deploy.yml. See deploy/README.md for the first-deploy walkthrough.
ISC