AI & product
Agent orchestration system
Eighteen specialised agents that build, research, test and operate, working as one system across two machines.
- My role
- Architect & builder
- Context
- My own infrastructure
My contribution
- Designed and built an orchestration system for 18 specialized agents across a Windows workstation and Ubuntu VPS.
- Implemented model fallback chains, rate limiting, Cloudflare Workers deployment and SSH tunneling.
Context
One general model doing everything is hard to trust and harder to control. Real work splits into specialities, and each one needs its own tools, its own limits and a plan for when a model fails.
What I did
I built a distributed system of 18 specialised agents covering development, research, QA and operations, running across a Windows workstation and an Ubuntu VPS. It has model fallback chains, rate limiting, deployment to Cloudflare Workers and SSH tunnelling between the machines.
Result
A structure where every agent has one job, a model that fails hands over to the next in its chain, and two machines behave as one system.
- specialised agents
- 18
- areas: development, research, QA, operations
- 4
- machines: a Windows workstation and an Ubuntu VPS
- 2
Built with
- Cloudflare Workers
- Ubuntu
- Windows
- SSH