After customising my Omarchy setup, I wanted to be able to reproduce it on
another computer without remembering every application, setting and workaround.
I built Wakiita to help turn those choices into a setup I could keep in Git.
Wakiita is an open-source agent plugin with three workflows: learning a setup,
capturing later changes, and restoring it. It helps create a personal repository
of recipes, each describing how to install, configure and verify one component.
Shared preferences can be kept separate from machine-specific settings, and a
recipe can have different installation routes for different operating systems.
The engineering challenge is deciding which work to leave to an agent and which
parts need explicit rules. Agents inspect the machine, interpret configuration
and carry out the setup. Python tooling validates the recipe manifests, selects
the applicable routes and machine overlays, and builds a dependency graph.
The agent then follows the plan, checking each component before its dependants
can proceed. Failed prerequisites leave dependent work blocked.
I tested the workflows through isolated capture-and-restore rehearsals with
fresh agents, alongside deterministic tests for the planning tools.
The rehearsals check that intended preferences survive restoration,
unrelated settings are preserved, and failures are reported accurately.
I use Wakiita to maintain my own setup. Its initial focus is Omarchy, while the
planner also handles personal repositories containing Windows recipes.
Source code
·
User manual