plato-core 0.1.0

Foundation types and mesh registry for the SuperInstance ecosystem

plato-core

Foundation types and mesh registry for the SuperInstance ecosystem.

Zero external dependencies. Everything else builds on top.

What's Here

• Types — TrainingTile, TileType, TileLifecycle, LamportClock, TrainingConfig, TrainingMetrics, AdapterConfig, content_hash()
• Mesh Registry — Auto-discovery of ecosystem packages via Python entry_points

Install

pip install -e .

Usage

Types

from plato_core import TrainingTile, TileType, content_hash, LamportClock

tile = TrainingTile(
    tile_id="chk-001",
    tile_type=TileType.CHECKPOINT,
    name="my-model-v1",
)

clock = LamportClock()
tile.lamport = clock.tick()

h = content_hash(b"some weights data")

Mesh Registry

The registry auto-discovers all installed SuperInstance packages:

from plato_core import MeshRegistry

registry = MeshRegistry()
matchers = registry.get_matchers()      # from plato-matcher
compressors = registry.get_compressors()  # from plato-compress
trainers = registry.get_trainers()       # from plato-training

Registering a Plugin

Any package can join the ecosystem:

1. Add plato-core as a dependency
2. Declare an entry point in pyproject.toml:

[project.entry-points."superinstance.plugins"]
my-plugin = "my_package._mesh:register"

3. Implement the register function:

# my_package/_mesh.py
def register(registry):
    registry.register("matchers", "my_matcher", lambda: MyMatcher)

Done. The mesh registry discovers it automatically.

Architecture

plato-core (this)       ← base types + registry
  ├── plato-training    ← training rooms, micro models
  ├── plato-matcher     ← semantic matching
  ├── plato-compress    ← compression/quantization
  └── ...               ← any future SuperInstance package

Each package is standalone. They communicate through shared types from plato-core and auto-discover each other via the mesh registry.

Development

pip install -e ".[dev]"
pytest tests/ -v

License

MIT