🎯 Syft Objects

Discover and use files you can't read. Syft Objects enables federated data science by letting you write code against mock data that runs on real data elsewhere.

Why Syft Objects?

In federated computing, you often need to:

• 🔍 Discover files that exist on other machines
• 📍 Address those files in your code
• 🧪 Test your code locally before running remotely
• 🔐 Control who can see what

Syft Objects solves this by providing a simple mock/private pattern for any file type.

Installation

pip install syft-objects

Quick Start

import syft_objects as so

# Create an object with mock (public) and private versions
analysis = so.create_object(
    name="Q4 Sales Analysis",
    mock_contents="Revenue up 10% to $2.5M",         # What others see
    private_contents="Revenue: $2,547,891.23",       # What you see
    discovery_read=["public"],                       # Who can discover it
    mock_read=["team@customer_company.com"],         # Who can see mock
    private_read=["cfo@our_company.com"]             # Who can see real data
)

# Browse available objects
so.objects  # Interactive table with search, filter, and permissions

Core Concepts

1. Every Object Has Two Versions

• Mock: Sample/demo data anyone can code against
• Private: Real data only authorized users can access

2. Permissioned Discovery

# Others can discover your object exists, even if they can't read it
so.objects  # Shows objects YOU can discover (not everyone sees the same list)

3. Write Once, Run Anywhere

# Test locally with mock data
netflix_data = so.objects["0b2f982d-6f82-48f3-b32e-3005e186e1cc"]
result = len(netflix_data.mock.obj)  # Works with mock

# Same code runs on private data elsewhere
def analyze_netflix_data():
    data = pd.read_csv(netflix_data.private.path)  # Runs on real data
    return len(data)

Common Workflows

Creating Objects from Files

# From existing files
dataset = so.create_object(
    name="Customer Dataset",
    mock_file="sample_100.csv",
    private_file="customers_full.csv"
)

# From folders
model = so.create_object(
    name="ML Model",
    mock_folder="model_demo/",
    private_folder="model_prod/"
)

Finding and Using Objects

# 1. Browse available objects
so.objects

# 2. Search and select (checkbox) objects of interest
# 3. Click "Python" button to copy code
# 4. Paste and use:
results = [so.objects["uid-here"]]
mock_data = results[0].mock.obj

Creating Derivative Objects

# Your analysis creates new syft objects
def my_analysis():
    # Load input
    input_data = pd.read_csv(input_path)
    
    # Process
    result = len(input_data)
    
    # Save as new syft object
    so.create_object(
        name=f"Row count: {result}",
        private_file="result.txt",
        mock_file="result_mock.txt",
        discovery_read=["researcher@university.edu"]
    )

CRUD Operations

Create

obj = so.create_object(name="My Object", mock_contents="demo", private_contents="real")

Read

obj.get_name()
obj.get_permissions() 
obj.mock.obj  # Access mock data

Update

obj.set_name("Updated Name")
obj.set_permissions(mock_read=["new@user.com"])

Delete

obj.delete_obj()

Web Interface

./run.sh  # Starts web UI at http://localhost:8004

Features:

• Search and filter objects
• Drag-and-drop file upload
• Edit permissions
• View/edit mock and private files

How It Works

Syft Objects creates YAML config files that point to mock and private versions of your data:

name: Sales Data
mock_url: syft://user@email.com/public/objects/sales_mock.csv
private_url: syft://user@email.com/private/objects/sales_real.csv
mock_permissions: [public]
private_permissions: [cfo@company.com]

The library handles all the complexity of:

• File management and syft:// URLs
• Permission checking
• Mock/private switching
• Cross-datasite discovery

Use Cases

• Federated Data Science: Write analysis code against mock data, run on real data elsewhere
• Privacy-Preserving ML: Train models on distributed data without seeing it
• Secure Collaboration: Share data structure without revealing contents
• Intermediate Results: Track outputs from multi-step federated computations

Learn More

• Tutorial: See 5mins.ipynb for hands-on examples
• Federated Execution: Use with syft-queue or syft-data-science
• Issues: https://github.com/OpenMined/syft-objects/issues

License

Apache License 2.0