Build with clarity.

01 / GETTING STARTED

Install from source.

Start with Python 3.9–3.12 and an isolated virtual environment. NumPy and Matplotlib power the core CPU workflow; no account or API key is required.

git clone https://github.com/krishanth7/EveryO.git
cd EveryO
python -m venv .venv
source .venv/bin/activate
python -m pip install -e .
everyo doctor
everyo demo

git clone https://github.com/krishanth7/EveryO.git
cd EveryO
python -m venv .venv
.venv\Scripts\Activate.ps1
python -m pip install -e .
everyo doctor
everyo demo

python -m pip install -e ".[onnx]"
# Optional numerical reference:
python -m pip install -e ".[tensorflow]"

02 / THE NUMERICAL FOUNDATION

Think in tensors.

A tensor holds numerical data, its shape, and gradient state. Python lists use float32 by default; NumPy arrays retain their precision.

import everyo as eo

x = eo.tensor([[1.0, 2.0]])
y = eo.tensor([[3.0], [4.0]])

print(x.shape)        # (1, 2)
print(eo.matmul(x, y))
# Tensor([[11.]], shape=(1, 1), dtype=float32)

03 / FOLLOW THE GRAPH

Meet your autograd engine.

EveryO records the multiplication because x requires gradients. The backward pass fills x.grad with [4.]: the derivative of x² at x = 2.

import everyo as eo

x = eo.tensor([2.0], requires_grad=True)
y = x * x
y.backward()

print(x.grad)  # [4.]

04 / COMPOSE LAYERS

A network you can inspect.

Connect a four-feature input to a three-class output with a hidden layer and ReLU activation. This model has 131 trainable parameters.

import everyo as eo

model = eo.Sequential(
    eo.Linear(4, 16, seed=0),
    eo.ReLU(),
    eo.Linear(16, 3, seed=1),
)

print(model.num_parameters())  # 131

05 / A COMPLETE EXPERIMENT

Generate data. Train. Evaluate.

This self-contained example creates a seeded dataset, holds out 20% for evaluation, and trains with Adam. Run it after installing EveryO.

import everyo as eo
from everyo.datasets import make_blobs

features, labels = make_blobs(
    n_samples=600, n_features=4, centers=3, seed=0
)
x_train, x_test, y_train, y_test = eo.stratified_split(
    features, labels, test_size=0.2, seed=0
)
model = eo.Sequential(
    eo.Linear(4, 16, seed=0), eo.ReLU(),
    eo.Linear(16, 3, seed=1),
)
trainer = eo.Trainer(
    model, eo.Adam(model.parameters(), lr=0.01),
    eo.CrossEntropyLoss(), metrics=["accuracy"],
)
train_loader = eo.DataLoader(
    eo.ArrayDataset(x_train, y_train),
    batch_size=32, shuffle=True, seed=0,
)
test_loader = eo.DataLoader(
    eo.ArrayDataset(x_test, y_test), batch_size=64,
)
history = trainer.fit(
    train_loader, epochs=20, validation_loader=test_loader
)
print(trainer.evaluate(test_loader))

Accuracy depends on the generated data and training run. Inspect the returned evaluation metrics rather than assuming a fixed result.

06 / KEEP YOUR WORK

Save the model. Keep the experiment.

EveryO stores model architecture and weights in a pickle-free .evo archive. After training the model above, save it and reload it for inference.

from pathlib import Path

Path("artifacts").mkdir(exist_ok=True)
eo.save(model, "artifacts/model.evo")
restored = eo.load("artifacts/model.evo")
restored.eval()

with eo.no_grad():
    predictions = restored(eo.tensor(x_test[:3]))

print(predictions.shape)  # (3, 3)

For supported ONNX exports, validate the exported graph in your target runtime. The traced exporter specializes control flow to the recorded execution.

07 / API REFERENCE

The essential API.

08 / WHEN SOMETHING DOESN’T CLICK

A shorter path to the fix.

Start with everyo doctor, then check the environment, tensor shapes, and optional integrations.

09 / KEEP EXPLORING

Go one level deeper.