How I Used Claude + OpenClaw to Migrate and Rebuild a Full-Stack Website in One Day

Author: Yao Song · NexAgent AI Solutions Tools: Claude Code (Opus 4.6) + OpenClaw Agent

Today I completed what would traditionally take one to two weeks: merging two independent server projects into a unified full-stack platform with a complete AI-driven operations system.

Background

I had two independently running projects:

NexAgent — AI automation agency website (Next.js + Tailwind + PM2)
OpenClaw Club — AI Agent hands-on course platform (Next.js + Flask + PostgreSQL + Docker)

They ran on different servers with different tech stacks and designs. Goal: merge into one unified platform.

Step 1: Data Migration (~10 min)

Claude Code operated both servers via SSH, automatically handling:

Project file packaging and transfer
PostgreSQL database export/import (including pgvector data)
AI Agent workspace (12 skills, memory data) migration
Docker Compose adaptation for new environment

Step 2: Backend Deployment (~15 min)

Docker Compose deployment of the full backend stack:

PostgreSQL 17 + pgvector — course data + vector search
Flask API — course management, blog CRUD, user auth
Redis — caching
MinIO — file storage
Discord Bot — community moderation
Memory Service — AI memory system (FastAPI + bge-m3 embeddings)

Step 3: Frontend Code-Level Merge (~30 min)

The most challenging part — merging two Next.js apps into one.

Route Design

Final route structure for nextagent.ca:

/zh, /en — Main site (bilingual)
/zh/openclaw, /en/openclaw — Course platform
/zh/blog, /en/blog — AI News Hub
/api/openclaw/* — Flask backend proxy

Key Technical Decisions

Next.js Route Groups with [locale] for site-wide i18n
Middleware routing for automatic language detection
API Proxy catch-all route forwarding to Flask
Style unification — deleted 2800+ lines of CSS, switched to Tailwind + shadcn/ui
OpenClaw Sidebar — fixed desktop sidebar + mobile bottom tabs

Step 4: AI Agent Migration

Full migration of OpenClaw Agent (codename: Ling Xiao):

13 Skills — course management, blog, site ops, memory, reminders, full-stack dev
Memory data — structured facts + vector memories + reminders
Bot API driven — no admin panel needed, Agent manages all content via API

Step 5: Automated Blog System

Built a daily auto-running AI news hub:

Sources: TechCrunch, The Verge, GitHub Trending, HuggingFace, Chinese AI outlets

Workflow:

Daily 6AM cron trigger
Fetch latest AI news from all sources
AI generates bilingual summaries + categorization + tags
Vector embeddings stored in pgvector for semantic search
Articles saved as drafts for review

Step 6: System Optimization

8GB RAM server — every megabyte counts:

Eliminated duplicate processes (~1GB saved)
Reduced Gunicorn workers (~110MB saved)
Docker container memory limits
systemd service management
Nginx direct Flask proxy (bypass Next.js layer)

Final: 2.5GB / 7.4GB used, all services healthy.

Key Takeaways

Claude Code Advantages

Direct SSH operations — no local clone needed
Self-healing errors — TypeScript errors auto-detected and fixed
Cross-server coordination — simultaneous multi-server operations
Full-stack coverage — frontend to backend to database to DevOps

OpenClaw Agent Value

Skill-based architecture — each capability is standalone, composable, extensible
Persistent memory — vector search + structured facts across sessions
Bot API driven — AI is the CMS, no traditional admin panel needed
Automated workflows — cron + AI + auto-categorization = hands-free operations

One Day's Work

2 servers merged
3 systems integrated (main site + courses + blog)
Unified Tailwind + shadcn/ui design
Bilingual support (EN/ZH)
AI memory system deployed
Automated news fetching
13 Agent Skills configured
System performance optimized

Interested in AI Agent development? Join our free open classes (every Friday in April, 8PM ET / 5PM PT) or visit nextagent.ca/openclaw.

Written by Yao Song, assisted by Claude Code + OpenClaw Agent.