Ultralytics Platform Quickstart#

Ultralytics Platform provides a guided workflow to upload datasets, train new YOLO models from pretrained weights, test completed models in the browser, and configure dedicated inference endpoints.

The following interactive diagram outlines the four primary stages of the Ultralytics Platform workflow. Click any stage or sub-step to access detailed instructions for that section.

graph LR
    A(Sign Up):::start --> B(Prepare Data):::proc --> C(Train):::proc --> D(Deploy):::out
    A -.- A1["<a href='#get-started'>Create account</a><br/><a href='#region-selection'>Select region</a>"]:::proc
    B -.- B1["<a href='#upload-your-first-dataset'>Upload dataset</a><br/><a href='#create-your-first-project'>Create Project</a>"]:::proc
    C -.- C1["<a href='#training-configuration'>Configure training</a><br/><a href='#monitor-training'>Monitor progress</a>"]:::proc
    D -.- D1["<a href='#test-your-model'>Test model</a><br/><a href='#deploy-to-production'>Deploy endpoint</a>"]:::proc

    click A "#get-started"
    click B "#upload-your-first-dataset"
    click C "#train-your-first-model"
    click D "#deploy-to-production"

    classDef start fill:#4CAF50,color:#fff
    classDef proc fill:#2196F3,color:#fff
    classDef out fill:#9C27B0,color:#fff

Get Started#

Ultralytics Platform offers a variety of easy signup options. You can register and log in using your Google or GitHub accounts, or with your email address.

Region Selection#

During onboarding, you'll be asked to select your data region. The Platform automatically measures latency to each region and recommends the closest one. This choice determines where your datasets, models, and managed training data are stored. Dedicated endpoint regions are selected separately when you deploy a model.

Free Credits#

Every new account receives free credits for cloud GPU training:

Complete Your Profile#

The onboarding flow guides you through three steps:

1. Profile - Enter your display name, unique username (permanent, cannot be changed later), organization (optional), and primary use case
2. Data Region - Select US, EU, or AP with a visual world map showing latency
3. Complete - Review your selections, optionally apply a promo code, and finish signup to claim your welcome credits

Home Dashboard#

After signing in, you will be directed to the Home page of Ultralytics Platform, which provides a welcome card with workspace stats, an overview video, quick access to datasets, projects, and storage, and a recent activity feed.

Sidebar Navigation#

The sidebar provides access to all Platform sections:

Each content section lists your five most recent items with a link to the rest. Hovering an item reveals a delete action that moves it to Trash; Trash itself is reached from search (Cmd+K) or by opening platform.ultralytics.com/trash.

Welcome Card#

The welcome card shows your profile, plan badge (which links to plan comparison), and workspace statistics at a glance. Each stat links to the matching workspace view:

Quick Actions#

Below the welcome card, the dashboard shows three cards:

• Datasets: Create a new dataset or drop images, videos, or dataset files to upload. Shows your recent datasets.
• Projects: Create a new project or drop .pt model files to upload. Shows your recent projects.
• Storage: Overview of your storage usage (datasets, models, exports) with plan limits.

A Recent Activity table at the bottom shows your latest datasets, projects, and deployments with their status and last update time.

Global Search#

Press Cmd+K (Mac) or Ctrl+K (Windows/Linux) to open the search bar. Search across pages, projects, datasets, and deployments instantly.

AI Chat Assistant#

A floating chat widget is available on every page. Click it to ask questions about YOLO training, annotation, deployment, or any Platform feature. The assistant provides context-aware help based on the current page.

Onboarding Tours#

The Platform includes guided tours that introduce key features as you explore different sections:

Restart Tours#

To replay any tour:

• Redo Tour button — Click your profile avatar (bottom-left of the sidebar) to open the account menu, then select Redo Tour. This resets all tours so they replay on your next visit to each section.
• URL parameter — Append ?tour= with the tour ID to restart one directly: ?tour=nav on the Home page, ?tour=project on a project page, or ?tour=dataset on a dataset page.

Upload Your First Dataset#

Open Annotate in the sidebar and click the + to create a new dataset. You can also drag and drop files directly onto the Datasets card on the Home dashboard.

The New Dataset dialog offers four sources. This quickstart uses Upload; the others are covered in Datasets and Integrations:

Uploads support multiple formats (full details in Datasets):

graph LR
    A[Drop Files]:::start --> B[Auto-Package ZIP]:::proc
    B --> C[Upload to Storage]:::proc
    C --> D[Process Data]:::proc
    D --> E[Resize & Thumbnail]:::proc
    E --> F[Parse Labels]:::proc
    F --> G[Compute Statistics]:::proc
    G --> H[Dataset Ready]:::out

    classDef start fill:#4CAF50,color:#fff
    classDef proc fill:#2196F3,color:#fff
    classDef out fill:#9C27B0,color:#fff

After upload, the platform automatically processes your data:

1. Images larger than 4096px are resized (preserving aspect ratio)
2. 256px thumbnails are generated for fast browsing
3. YOLO, COCO, and Ultralytics NDJSON labels are parsed and validated
4. Statistics are computed (class distribution, heatmaps, dimensions)

my-dataset.zip
├── data.yaml          # Class names and splits
├── train/
│   ├── images/
│   │   ├── img001.jpg
│   │   └── img002.jpg
│   └── labels/
│       ├── img001.txt
│       └── img002.txt
└── val/
    ├── images/
    └── labels/

Read more about datasets and supported formats for detect, segment, pose, OBB, and classify.

Create Your First Project#

Projects help you organize related models and experiments. Open Train in the sidebar and click the + to create a project. You can also drop .pt weights onto the Projects card on the Home dashboard to create a project and import them in one step.

Enter a name and optional description. Projects organize model runs and imported or cloned model weights, with charts for comparing completed training results.

Read more about projects.

Train Your First Model#

From your project, click New Model to open the Train New Model dialog. You can also start from a dataset page, in which case the dataset is locked in and you pick the destination project instead.

Training Configuration#

1. Base Model: Select official Ultralytics weights or one of your own trained models. The dialog warns if the model task and dataset task don't match.
2. Dataset: Choose a ready dataset with at least one train image, at least one validation or test image, and at least one labeled image
3. Parameters: Set Epochs (default 100), Batch Size (blank means auto), Image Size (default 640), and an optional run Name. Expand Advanced Settings to edit any other Ultralytics training argument in a YAML editor.
4. Select GPU: On the Cloud Training tab, choose compute based on your budget and model size. The default is RTX PRO 6000 (96 GB Blackwell, $2.09/hr), which handles every YOLO26 variant. See the full GPU pricing table or the Cloud Training GPU step for the complete list and tier gating.
5. Start Training: Review the estimated cost and duration next to your credit balance, then click Start Training.

Monitor Training#

Once training starts, open the model's Train tab to monitor progress in real time through three subtabs:

Metrics are streamed in real-time via SSE (Server-Sent Events). Once validation artifacts exist, the Charts subtab splits into Training and Validation views, with the confusion matrix, PR curves, and F1 curves under Validation.

Read more about cloud training.

Test Your Model#

After training completes, test your model directly in the browser:

1. Navigate to your model's Predict tab
2. Upload an image, drag and drop, capture one from your webcam, or click an example image (auto-inference on drop)
3. View task-appropriate prediction overlays, per-stage timings (preprocess, inference, postprocess, network), and the raw JSON response

Adjust inference parameters:

Under API Docs, the Predict tab shows example code in Python, JavaScript, and cURL, pre-filled with the parameters you selected above. Deploy the model first, then replace the placeholder URL and key with the values from your endpoint:

import requests

url = "https://your-deployment-url.run.app/predict"
api_key = "YOUR_API_KEY"
args = {"conf": 0.25, "iou": 0.7, "imgsz": 640}

with open("image.jpg", "rb") as f:
    response = requests.post(
        url,
        headers={"Authorization": f"Bearer {api_key}"},
        data=args,
        files={"file": f},
    )

print(response.json())

Read more about inference.

Deploy to Production#

Deploy your model to a dedicated endpoint for production use:

1. Navigate to your model's Deploy tab
2. The Platform measures latency to all 42 regions and plots them on a world map, colored on a green-to-red gradient (lower latency is greener, higher latency is redder)
3. In the Region Latency table below the map — sorted by latency from your location — find the region you want. Use Rescan to re-measure at any time.
4. Click Deploy on that row to create your endpoint

graph LR
    A[Select Region]:::start --> B[Deploy]:::proc
    B --> C[Provisioning]:::proc
    C --> D[Running]:::out
    D --> E{Lifecycle}:::decide
    E --> F[Stop]:::error
    E --> G[Delete]:::error
    F --> H[Resume]:::proc
    H --> D

    classDef start fill:#4CAF50,color:#fff
    classDef proc fill:#2196F3,color:#fff
    classDef decide fill:#FF9800,color:#fff
    classDef out fill:#9C27B0,color:#fff
    classDef error fill:#F44336,color:#fff

Once provisioning completes, your endpoint provides:

• Unique URL: HTTPS endpoint for API calls
• Scale-to-zero behavior: Idle endpoints scale to zero (deployments currently run a single active instance)
• Monitoring: Request metrics and logs

After deployment, you can manage all your endpoints from the Deploy section in the sidebar. The deployments dashboard shows a global map with your active deployments, 24-hour metrics (total requests, active deployments, error rate, and P95 latency), and a list of every endpoint.

Read more about endpoints.

Remote Training (Optional)#

If you prefer to train on your own hardware, you can use your API key to train anywhere and stream metrics to Ultralytics Platform.

The fastest route is the Local Training tab in the training dialog: it builds the full command for your selected model, dataset, and parameters, and fills in an API key (creating one if you don't have one yet). Copy it and run it in your terminal.

To assemble the command yourself:

1. Generate an API key in Settings > API Keys
2. Set the environment variable and train with a username/project value for project:

export ULTRALYTICS_API_KEY="YOUR_API_KEY"

yolo train model=yolo26n.pt data=coco.yaml epochs=100 project=username/my-project name=exp1

Read more about API keys, dataset URIs, and remote training.

Feedback & Help#

The Help page in the sidebar footer collects documentation links, video walkthroughs, and community resources, and includes an in-app feedback form. You can rate your experience from 1 to 5 stars, choose a feedback type (Bug, Feature, or General), and attach a screenshot.

If you need more help:

• AI Chat: Click the floating chat widget on any page for instant help
• Documentation: Browse these docs for detailed guides on datasets, annotation, training, deployment, and billing
• Discord: Join our Discord community for discussions
• GitHub: Report issues on GitHub
• REST API and Python SDK: See the API reference, install the ultralytics-platform client, or try the interactive API docs for programmatic access to all Platform features