Introducing DLWΛY
Deep learning,
your way.
A professional-grade ML workspace that lives in your browser. Write code, design architectures, train on GPU, ship models — without ever touching infrastructure.
01 — Ingest
Everything
starts as noise.
Pull a dataset from anywhere and it lands in the same workspace — typed, profiled, and ready to inspect. No loaders to write, no schemas to hand-roll.
- CSV
- Parquet
- JSONL
- Kaggle
- Hugging Face
02 — Prepare
Then you
give it order.
Seventy-plus canonical transforms, composed into a pipeline you can read top to bottom. Every step is versioned, reversible, and reproducible.
- 01DropDuplicatessubset: all
- 02ImputeNumericstrategy: median
- 03StandardScalerwith_mean: true
- 04OneHotEncodermin_freq: 0.01
- 05TrainTestSplit0.8 / 0.2
03 — Architect
Build the
network, layer
by layer.
Drag, drop, connect. Shapes are inferred live as you go, so a mismatch is caught on the canvas instead of forty minutes into a training run.
- Conv2D64 × 3×332×32×64
- BatchNorm—32×32×64
- MaxPool2D2×216×16×64
- Dense256, relu256
- Dense10, softmax10
04 — Train
Watch it
learn, live.
Training runs on your own GPU, right in the tab. Loss and accuracy stream back in real time, next to the architecture that is producing them.
05 — Deploy
Then send it
into the world.
One trained model, many destinations. Try it in the playground, then export a quantized artifact, a browser bundle or a server wrapper — from the same run.
POST http://localhost:8080/v1/resnet-cifar-42/predict- Playground
- Quantized model
- Browser bundle
- Server wrapper
Start building,
your way.
Free to start. No credit card, nothing to install, no infrastructure to babysit.
See everything insideWhat it is
One workspace for the whole machine learning workflow
DLWAY (written DLWΛY, short for “Deep Learning, Your Way”) is a machine learning workspace that runs entirely in your browser. Load a dataset, clean it, design a neural network, train it on your own GPU, chart the results and export the model, without leaving the tab and without installing anything.
The work usually spread across a notebook, a visual pipeline tool, a dashboard product and a couple of dataset sites sits in one project here. Each module reads what the others produce, so data loaded once is ready for preparation, training and charting.
- installs, drivers or environments to set up
- 0
- data-preparation transforms in PrepFlow
- 70+
- database engines you can query live
- 7
- runtimes in the tab: Python, JavaScript and TensorFlow.js
- 3
Inside the Studio
Every stage has its own tool, and they share one project
Use only the code editor, only the visual tools, or move between them. A model drawn on the canvas becomes code you can edit; a dataset cleaned in PrepFlow is there to train on.
Explorer
A full code editor built on Monaco, the engine behind VS Code. Run Python and JavaScript in the tab, with a terminal, project-wide search and a git client beside your files.
Read the guideModel Builder
Design neural networks by dragging layers onto a canvas. Tensor shapes and parameter counts update as you connect them, and the result exports as TensorFlow.js, Keras or PyTorch code.
Read the guideData Hub
Load files and folders, connect PostgreSQL, MySQL, MongoDB, Snowflake, BigQuery and other databases, or pull datasets from Kaggle and Hugging Face. Every dataset is profiled and versioned.
Read the guidePrepFlow
Clean and reshape data as a graph you can read: imputation, encoding, scaling, outlier handling and resampling, chosen from more than seventy transforms instead of hand-written pandas.
Read the guidePipelines
Chain data, preparation, training and reporting steps into a run you can repeat, schedule, or trigger when a dataset changes, with a history of every run.
Read the guideDashboard
Build charts by dropping fields onto rows, columns and marks, add calculated fields and parameters, and watch training metrics arrive live.
Read the guideResearch
Start from a research paper. DLWΛY reads the PDF, shows you what it found, and generates an editable project with code, runs and figures.
Read the guideFine-Tune
Browse Hugging Face models, bind one of your datasets, and train in the browser or generate a fine-tuning script to run elsewhere.
Read the guideDeployment
Try a trained model in the playground, then export it as a quantized artifact, a browser bundle or a server wrapper.
Read the guideAI Copilot
An assistant that works inside your project. Ask it about your code, or let agent mode read files, edit them and run Python, asking before it changes anything.
Read the guide
Local first
It runs on your machine, so your data stays there
Your data stays on your device
Parsing, SQL, data preparation, training and charting all run in your browser, in Web Workers and WebAssembly. A dataset is only sent anywhere when you connect a database, import from a hub, attach a remote kernel or call an AI provider.
Nothing to install
Open the Studio and start. There is no Python environment to set up, no GPU driver to match, and no installer to get approved on a managed or shared machine.
Your hardware, no meter running
Models train on the GPU you already have, through WebGPU with WebGL as the fallback. There is no cloud session to time out halfway through a run.
The details are in Privacy and storage and the privacy policy.
Who it is for
Built for people who want to get to the model
Students and educators
A complete machine learning workflow on any lab machine or laptop, without cloud credits or software installs.
Researchers
Reproduce a paper as an editable project, keep every run and its configuration, and export the lot as one archive.
Teams with sensitive data
Work on financial, medical or otherwise regulated data that is not allowed to leave your own machines.
Practitioners on a budget
Code, visual pipelines, dashboards and dataset hubs in one tab, instead of a subscription for each.
New to machine learning? Start with the tutorials. Wondering about cost? See pricing.
What is DLWAY?
DLWAY (written DLWΛY, short for Deep Learning, Your Way) is a machine learning and data workspace that runs in your web browser. It brings a code editor, a visual neural network builder, a data hub, data preparation, pipelines, dashboards and model export together in one tab.
Do I need to install anything?
No. DLWAY runs in a current desktop browser. Python runs through Pyodide, which is CPython compiled to WebAssembly, and models train with TensorFlow.js, so there is no environment, driver or package to install.
Is DLWAY free?
Yes. Everything in DLWAY is free to use, with no paid plans and no credit card. That may change in the future; if it does, the pricing page will say so. Services you connect, such as an AI provider with your own API key, set their own prices.
Is my data uploaded to your servers?
No. Datasets are parsed, queried, prepared and trained on inside your own browser, and projects are stored on your device. Data leaves it only when you choose to: connecting a database, importing from Kaggle or Hugging Face, attaching a remote kernel, or using an AI provider.
Do I need a GPU?
No. Training uses whatever GPU your browser can reach through WebGPU, falls back to WebGL, and still works on a machine with no dedicated GPU, only more slowly.
Which browsers does DLWAY support?
Chrome and Edge 113 or later are recommended because they support WebGPU. Other current browsers work with WebGL acceleration.
Can I use Python and PyTorch?
Python runs in the browser with libraries such as NumPy, pandas and scikit-learn. The Model Builder exports PyTorch, Keras and TensorFlow.js code, and you can attach a Jupyter or Kaggle kernel to run native PyTorch or TensorFlow.
How is DLWAY different from a cloud notebook?
A cloud notebook runs your code on someone else's machine. DLWAY runs it on yours, so there is no session limit and nothing to upload, and it adds visual tools for building models, preparing data and charting results alongside the code.
Open a tab. Train a model.
Free, with no credit card and nothing to install.