Developers

Five minutes to a working example

GeometriX ships as Python packages on PyPI today. Pick your side of the platform, install, and run a real example. Python 3.12 required.

1 · Install

$ pip install modeva
# predictive AI: Model Studio + Model Assurance

$ pip install knowlytix
# agentic AI: Agent Studio, Agent Assurance, Governance, Knowledge Fabric

These two packages are the platform. modeva uses a free developer tier with signed-license activation; the knowlytix family is Apache-2.0.

2 · Run something real

Train and diagnose a model

from modeva import DataSet
from modeva.models import MoLGBMRegressor

ds = DataSet()
ds.load("BikeSharing")

model = MoLGBMRegressor(name="LGBM", max_depth=2,
                        n_estimators=100)
model.fit(ds.train_x, ds.train_y.ravel())

# then: test suite, explanations, comparisons
# ts.explain_pdp(features="hr"), tsc.compare_accuracy_table(...)

Run a governed agent

from knowlytix.harness import Harness, HarnessConfig

h = Harness(HarnessConfig(stores_dir="gms_stores"))
h.register_gms_tools()
h.register_modeva_tools()

response = h.run(
    "Load BikeSharing, train XGB depth-2, "
    "run FANOVA, diagnose accuracy")
h.save()

3 · Which product lives in which package

Every GeometriX product ships today inside one of two installable packages.

GeometriX product	Package today	Where to start
Model Studio	modeva	from modeva import DataSet · modeva.models
Model Assurance	modeva	TestSuite APIs — ts.diagnose_, ts.explain_, tsc.compare_*
Agent Studio	knowlytix	from knowlytix.harness import Harness
Agent Assurance	knowlytix	knowlytix.harness.testing · knowlytix.benchmark
Governance	knowlytix	knowlytix.harness.governance
Knowledge Fabric	knowlytix	knowlytix.knowledge · knowlytix.kal

4 · Go deeper

Documentation

User guides and full API reference for the predictive stack; package documentation for the agentic stack. Unified GeometriX docs are consolidating here.

modeva.ai docs →

Runnable examples

Published notebooks: BikeSharing and SimuCredit (high-code and low-code), plotting, and a California Housing pipeline — runnable in Colab.

Example notebooks →

Benchmarks

FinStructBench ships in the box: auto-generated questions from document graphs, scored against graph-verified ground truth, with a CLI.

Research & benchmarks →