MYLO documentation

Overview

MYLO is a browser control plane for designing, editing, analyzing, and optimizing neural networks. The web app never runs ML workloads itself. Every heavy operation, loading models, CSV analysis, architecture edits, training, export, runs on an agent.

You must be Connected to an agent before you can open or create projects. How to connect to an agent is described below.

Getting started

1. Open the landing page and choose Get started, or go to /auth.
2. Sign in with GitHub or Google.
3. You land on Home — the hub before the IDE.

The connection indicator shows Connected, Connecting, or Offline. Connect a client before opening a project.

Local agents

A local agent lets you keep models and compute on a machine in your own network.

1. Get the Desktop Agent:

Clone the MYLO Desktop Agent repository:

2. Install dependencies

From the agent directory, create a virtual environment and install requirements:

cd MYLO-Desktop-Agent
python -m venv .venv
# Windows:
.venv\Scripts\activate
# macOS / Linux:
source .venv/bin/activate

pip install -r requirements.txt

3. Configure host and port

The agent reads .env. A typical local setup:

SERVER_IP="127.0.0.1"
SERVER_PORT="8080"

Use 0.0.0.0 only if you intentionally need LAN access. The web Clients form defaults to http://127.0.0.1:8080.

4. Start the agent

python App.py

Confirm it is up with a health check:

curl http://127.0.0.1:8080/
# → {"message": "MYLO AGENT is running!"}

5. Connect from the web app

1. Sign in at /auth, then open Home → Clients.
2. Click New config, choose type LOCAL.
3. Set a name (for example “My laptop”).
4. Address: http://127.0.0.1:8080 (default).
5. API key: any string you choose (the UI defaults to 1234). On first connect, the web app initializes the agent session with a hashed form of this key — keep the same key in the Clients config.
6. Save, then select the config. Status should become Connected.

Cloud agents (Currently Disabled)

Prefer not to install anything locally? Create a CLOUD client instead.

1. Home → Clients → New config.
2. Type: CLOUD, pick a name, save.
3. The app provisions a Docker agent via the orchestrator. You do not enter an address or API key — those are managed securely for you.
4. If the session goes offline, use Restart on the client card to spin up a fresh container.

Deleting a CLOUD config also tears down its container.

Projects & Catalog

Projects

• New Project — name it and open an empty Dashboard (requires Connected).
• Open on an active agent project — continue where you left off.
• Upload Project — load a .mylo file from disk into the agent.
• Cloud Projects — open projects saved to your account (downloaded into the agent, then Dashboard).

Catalog

Shared resources: full projects (.mylo) and standalone models (.pt2, .onnx). Open downloads the asset into your connected agent and opens the Dashboard (editor or visualiser, depending on type).

Dashboard

The Dashboard is the main IDE: a menu bar on top and dockable panels you can move, resize, and detach. The MYLO logo returns you to Home.

Project menu

View menu

Model Editor, Optimization, Diagnostics, and Visualiser also open from Project Manager actions.

Workflows

CSV and data analysis

1. Project → Load CSV (or upload from Project Manager).
2. Open Dataset Info for column stats, correlation matrix, and quality signals.
3. Use Data viewer for the raw rows. Feature and target columns feed Optimization later.

Visualiser

Read-only graph of layers and data flow. Click a node to inspect it in Properties. Edit architecture in the Model Editor instead.

Model Editor

• Insert or delete layers from the context menu
• Edit parameters in the side panel
• Connect compatible ports; invalid edges are rejected
• Undo / redo from the toolbar
• Save to validate and persist on the agent

Optimization

1. Open Optimization from a model slot in Project Manager.
2. Set input features, target column, problem type, epochs/generations, strategy (neuroevolution or Optuna), and encoding.
3. Start the run — progress streams live. When finished, review metrics, architecture diffs, and download the optimized model.

Diagnostics

Quick health check without a full optimize: forward pass, shapes, parameter count, and warnings (unused layers, dtype issues, and similar).

AI Assistant

Chat about the current project — describe the architecture, insert layers, ask about CSV correlations, or open panels. After AI edits, review them in the Model Editor and save if they look right.

Save and export