Skip to main content

One service

Declare it, run it, open it:
Edit files locally; tesser sync $BOX ships the delta and HMR picks it up (exec and dev sync implicitly). If you changed something the server can’t hot-absorb — env, config, deps — tesser sync $BOX --restart.

A mesh

Say the architecture is frontend → backend → worker. Each service gets a manifest; deps are the loopback ports the app already uses:
Pin the shared instances once (usually CI does this on every push to main):
Now work on backend in a worktree:
Nobody configured a URL. The frontend’s localhost:5001 reaches your backend; your backend’s localhost:5002 reaches the shared worker; the shared mesh never routes into your dev instances. Switch what any box’s deps point at — or what localhost:3000 shows — from the panel injected into every proxied page. Every box is also directly addressable, no selection involved:

Tests and heavy commands

Never on a serving box. exec without a box ID targets the worktree’s workbench — a synced box with nothing running on it, claimed from the warm pool on first use:

Walking away

Do nothing. Idle boxes sleep (workbenches after 10 minutes, instance boxes after 2 hours) at ~cents/day; boxes asleep 16 hours are removed. Any command targeting a sleeping box wakes it in ~40 seconds with caches warm. tesser rm $BOX when a worktree is done; tesser ls when you’ve lost track.