A 3-Stage Seni Bina Penghalaan Pintar for My Own Ejen AI✎ Edit

When I first joined this OpenClaw group, members actively shared experiments, technical experiences and projects.

Then came spam and scammers, followed by competing product promotions. And now, memes have practically taken over the group.

That drift is part of why I went the other direction and built my own lightweight Ejen AI to replace OpenClaw - no unnecessary complexity, no bloated function surface I never asked for.

Arsitektur Penghalaan Pintar 3 Peringkat Saya:

1. Redis & MySQL Cache – Reuse existing data and results so I'm not paying to recompute what I already have.
2. Frequent Scripts – Handle the repetitive work deterministically, with zero unnecessary LLM calls in the path.
3. Intelligent Model Penghalaan – Hit the lowest-cost capable model first and only escalate up the chain when the task actually demands it.

Timbunan Teknologi: Web-CLI, Redis, MySQL and Telegram as my on-the-go interface.

Running my own Ejen AI, I've shipped 100+ standalone web applications, along with a lot of server and infrastructure work underneath them:

• Linux server provisioning and VM management
• PostgreSQL and MySQL
• Redis caching
• Docker and container management
• Nginx and reverse proxy
• Gitea and Git version control
• Timeshift, snapshots and backup management
• Server monitoring and deployment automation

The interesting part? Total AI model costs stay below USD 40/bulan, and that includes running my own distilled model.

Application development, server operations and infrastructure management - all driven through my own Ejen AI.

Less overhead. Smarter routing. Lagi engineering.

How about everyone else? What have you been building, experimenting with or improving lately?