Welcome to the Docs

Step 1: Upload a Dataset

Navigate to Datasets in your project and upload your data. We support CSV and XLSX formats.

• Ensure your data has clear column headers
• Files must be under 500MB
• Automatic column type detection with manual overrides

Step 2: Create a Model

You have two options for creating a model:

import mldash

# Train your model as usual...
model.fit(X_train, y_train)

# Export to ML-Dash
mldash.export(
    model=model,
    name="my-model",
    target={"label": "categorical"},
)

!pip install mldash-sdk

import mldash
mldash.login()

model.fit(X_train, y_train)

mldash.export(
    model=model,
    name="my-model",
    target={"label": "categorical"},
)

Step 3: View Your Model

Once created, view your model in your project's models page:

• See training metrics and metadata
• Set visibility (Private / Public)
• Edit name, description, and tags
• View model changelog for all events

Step 4: Test Your Endpoint

Go to your project's API page to test predictions:

• Select a model and send a test prediction
• Copy the endpoint URL for external use
• Or use Dash AI to test via chat
• View Python, cURL, and JavaScript code examples

Uploading Datasets

The platform supports the following file formats:

• CSV: Comma-separated values (recommended)
• XLSX: Excel spreadsheets

Dataset Requirements

• Upload limit is 500 MB on all plans (Free, Plus, Pro); plans differ by total storage and other quotas
• Column headers are required in the first row
• At least 50 rows recommended for training
• Numeric features should be properly formatted
• Categorical features should be consistent (no typos)

Data Preprocessing

The platform uses sentence transformers for text processing:

• Text descriptions are converted to 384-dimensional embeddings
• Numerical features scaled with StandardScaler
• Categorical features one-hot encoded
• Train/test split configurable (default 80/20)

Dataset Management

Click on any dataset in the list to view its changelog. You can also:

• Refresh datasets with new data (same schema required)
• Download datasets for offline analysis
• Edit dataset names and descriptions
• Delete datasets no longer needed

Common Issues

Studio

Studio is an in-browser Python notebook for training models directly on the platform.

• Train any model (sklearn, XGBoost, LightGBM, CatBoost, TensorFlow)
• Export with mldash.export(model=model, name="...", target={...})
• Pre-built templates available for common workflows
• Full Python environment with popular ML libraries

SDK Export

Bring models trained in your own environment into ML-Dash using the SDK.

Supported Formats

The SDK automatically serializes your model. Supported frameworks for SDK export: scikit-learn, XGBoost, LightGBM, and CatBoost. Other frameworks (TensorFlow/Keras, PyTorch, statsmodels) can be imported via the web UI Import Wizard.

Using mldash.export()

!pip install mldash-sdk

import mldash
mldash.login()

# Train your model
model.fit(X_train, y_train)

# Export to ML-Dash
mldash.export(
    model=model,
    name="my-model",
    target={"label": "categorical"},
)

On your local machine, prefer uv tool install mldash-sdk or pipx install mldash-sdk for an isolated CLI install.

The SDK bundles your model with metadata and uploads it. Your model gets a REST endpoint automatically.

Authentication options

mldash.login() prompts interactively, but for Colab, scripts, or CI you can pass credentials directly or via environment variables. Resolution order: explicit args → environment variables → Colab Secrets → stored credentials file.

Explicit args (one-off scripts):

mldash.login(url="https://api.mldash.com", api_key="mldash_xxxxxxxx")
# or skip login entirely and pass on each call:
mldash.export(model=model, name="My Model",
    api_key="mldash_xxxxxxxx", project="my-project")

Environment variables (CI, shell scripts):

export MLDASH_URL=https://api.mldash.com
export MLDASH_API_KEY=mldash_xxxxxxxx
# Then in Python: no login() call needed
mldash.export(model=model, name="My Model")

Colab Secrets (recommended for Colab notebooks):

Click the 🔑 key icon in Colab's sidebar, add MLDASH_URL and MLDASH_API_KEY, and toggle Notebook access on. Then in your notebook:

import mldash
mldash.export(model=model, name="My Model")  # auto-resolves from Colab Secrets

Get an API key at /developers/keys. The default URL is https://api.mldash.com if not specified.

Supported Frameworks

ML-Dash supports models from these frameworks:

Training Metrics

Metrics are displayed when available (set during training or import).

Model Comparison

Compare multiple models side by side to find the best performer.

• Select 2 or more models from your project's models page
• Click "Compare" to open the comparison view
• Side-by-side metrics, hyperparameters diff, and training metadata

Explore & Sharing

Share your models and datasets with the community, or browse what others have built.

Visibility Controls

• Model visibility: Public/private toggle on your project's models page
• Dataset visibility: Public/private toggle on your project's datasets page
• Public items appear on the Explore page for all users

Explore Pages

• /explore — Browse public models
• /explore/datasets — Browse public datasets with schema preview and download
• /explore/saved — View your bookmarked models and datasets

Dash AI

An AI assistant that helps non-technical users interact with models through conversation.

Getting Started

• Access via /dash-ai or the navigation bar
• Select a trained model to chat about
• Conversations are persisted with auto-generated titles

Capabilities

• Predict: Describe your data in natural language and Dash AI extracts features and calls the model
• Explain: Ask "why?" to get SHAP-based explanations narrated naturally
• Inspect: Ask about model metrics, training data, project state, or compare models side-by-side
• Manage notebooks (Studio right panel): create, edit, and insert cells; open existing notebooks
• Clean and reshape data: clean datasets, filter rows, join datasets, fix column types
• Debug your environment: list packages installed in your Studio venv, look up current PyPI metadata, diagnose Python version issues
• Navigate: ask Dash AI to open Studio, model pages, or project views

Mentions & References

Type @in the chat to reference any model, dataset, project, or notebook by slug. Dash AI loads the resource's metadata into context automatically:

• @model/your-slug — pull in a model's metrics and schema
• @dataset/your-slug — preview rows and column types
• @project/your-slug — project summary and resources
• @notebook/your-slug — first cells of a Studio notebook

Dash Assist Beta

Local AI autocomplete in Studio. As you type, Dash Assist suggests the next few lines as greyed-out ghost text — powered by a local LLM running on your laptop, so no code ever leaves the machine.

How It Works

• Connect Studio with mldash connect — suggestions appear as you type
• Tab to accept, Esc to dismiss, or just keep typing past it
• Powered by Qwen2.5-Coder 1.5B via llama-cpp-python, running locally
• Uses fill-in-the-middle (FIM) prompting, so completions fit your surrounding code, not just continuations
• 300ms frontend debounce, 128-token output cap, FIM-boundary stop tokens for clean stops

Why Local

• Private: your code never leaves your laptop
• Fast: sub-second suggestions, no network round-trip
• Offline-friendly: works without an internet connection once the model is downloaded

Embeddable Playground

Embed a prediction form for any public model on your own site.

• URL pattern: /embed/{username}/{model-slug}
• Compact prediction form with no dashboard chrome
• Find the embed code snippet on Explore → model detail → API tab
• Use an <iframe> with configurable height

Notifications

Stay informed about model sharing and account activity.

• Dataset processed
• Model published / unpublished
• Rate limit warnings

Bell icon in the header shows unread count. Configure per-type preferences (in-app and email toggles) at /settings/notifications.

Overview

The Prediction API allows you to integrate your trained models into external applications. Send input data and receive predictions in real-time.

Authentication

All API requests require an API key. Create and manage keys in the API Keys section.

Rate Limiting & Subscription Plans

API requests are rate-limited based on your subscription plan. See pricing →

Prediction Endpoint

Make predictions with a trained model. The identifier can be a model slug or ID. Your own models are checked first, then public models.

Request

{
  "text": "Coffee at Starbucks $4.50",
  "features": {           // optional
    "amount": 4.50,
    "category": "food"
  }
}

Response

{
  "prediction": "Food & Dining",
  "probabilities": {      // classifiers only
    "Food & Dining": 0.87,
    "Shopping": 0.09,
    "Transport": 0.04
  },
  "model_id": 1,
  "model_slug": "transaction-categorizer",
  "model_name": "Transaction Categorizer",
  "model_type": "logistic_regression"
}

Code Examples

import requests
import os

API_KEY = os.getenv('ML_DASH_API_KEY')

response = requests.post(
    "https://api.mldash.com/api/v1/predict/my-model/",
    headers={
        'X-API-Key': API_KEY,
        'Content-Type': 'application/json'
    },
    json={"text": "Coffee at Starbucks"}
)

result = response.json()
print(f"Prediction: {result['prediction']}")

Error Responses