Lingot

The mint for the reserve currency of embodied AI. Lingot is the umbrella orchestrator of the Autarkis spatial stack — the thin layer that depends on the worker services and conducts them through one pipeline:

spec -> cadastre (registry + solve) -> chitin (physics .phys) -> oneiros (render) -> StagedScene

Lingot is intentionally thin. It owns the order and the data hand-off; the work lives in the workers, each an independently published package.

Install

pip install lingot            # Python orchestrator + CLI   (PyPI: lingot)
npm install @autarkis/lingot  # JS orchestrator / client    (npm bare `lingot` is taken -> scoped)

The Rust force-directed solver is not vendored here — it ships as its own worker package and is pulled in as a dependency. Lingot orchestrates; it does not implement.

Local development

The worker repos are siblings (../cadastre, ../chitin, ../oneiros). Wire them in editable mode, then run the pipeline on a spec:

just dev-install
just run path/to/spec.json

Every pipeline stage delegates to a worker adapter in src/lingot/workers.py. Stages with no wired worker raise a clear error naming what to connect — the skeleton runs end-to-end the moment the seams are filled.

Layout

lingot/
|- src/lingot/         # Python orchestrator
|  |- pipeline.py      #   stage order + data hand-off
|  |- workers.py       #   one adapter seam per worker service
|  \- cli.py           #   `lingot run spec.json`
|- ts/                 # @autarkis/lingot — JS orchestrator / client
\- justfile            # dev / test / build across py + ts

See ../LINGOT.md for the full thesis.