Wez: my own agent harness
What it is
My own agent harness, where a crew of AI agents works in my projects. Every request, tool call, file change, cost and approval is visible while it happens. It is how I build, and you get the same visibility in what I build for you.
The agents have roles such as architect, designer and QA, and the harness runs locally. It is where I build both client solutions and my own products.
How it’s built
An engine on Bun with the agent loop, providers, tools and SQLite state, written with Effect. The client is React, with a desktop app in Electrobun and mobile clients in Capacitor.
An MCP server lets Claude Code or Codex drive the crew. Worktrees and file leases let several agents work in parallel, machines join a fleet over Tailscale, and prompt caching is part of the design.
Stack: Bun, Effect, TypeScript, React, SQLite, Electrobun, Capacitor, MCP, Tailscale.
What it shows
Afterwards you can see what the agents did, what it cost and who approved it. I build that traceability into what I deliver.
- 1,300+
- commits
- 6
- agents in the crew
- 3
- clients: desktop, mobile and CLI
Private project, in daily useRelated service: AI developmentNext case: Lingo