Setting Up Your Local SyftBox Environment

curl -fsSL https://syftbox.net/install.sh | sh

https://www.syftbox.net/

syftbox --version

SyftBox version 0.1.0

syftbox client

ls -la ~/SyftBox

~/SyftBox/
├── .syftbox/              # Configuration and keys
│   ├── config.json        # Client configuration
│   └── client_config.json # Connection settings
├── datasites/             # Peer datasites appear here
├── public/                # Your public folder (readable by all)
├── private/               # Your private data (only you can read)
├── api_data/              # API and app data
└── sync/                  # Sync metadata

import syft_rds as sy

# Initialize as admin on your own datasite
do_email = "[email protected]"  # Your email
do_client = sy.init_session(host=do_email, email=do_email)

# Verify admin access
print(f"Admin access: {do_client.is_admin}")

from pathlib import Path

# Prepare your dataset
DATASET_DIR = Path("./my-dataset").absolute()

# Ensure it has the required structure:
# my-dataset/
# ├── private/
# │   ├── train.csv
# │   └── test.csv
# └── mock/
#     ├── train.csv
#     └── test.csv

# Create the dataset in SyftBox
do_client.dataset.create(
    name="pima-indians-diabetes-database",
    asset_path=DATASET_DIR,
    description="Diabetes prediction dataset",
)

print("✅ Dataset created successfully")

# List all datasets
datasets = do_client.dataset.get_all()
for ds in datasets:
    print(f"Dataset: {ds.name}")
    print(f"  Private path: {ds.get_private_path()}")
    print(f"  Mock path: {ds.get_mock_path()}")

import syft_rds as sy

# Initialize your session
ds_email = "[email protected]"
ds = sy.init_session(host=ds_email, email=ds_email)

print(f"Data scientist logged in: {ds_email}")

# Connect to first data owner
do1_email = "[email protected]"
do1_client = sy.init_session(
    host=do1_email,
    email=ds_email,  # Login as yourself (guest)
    start_syft_event_server=False,
)

print(f"Connected to {do1_email}")
print(f"Admin access: {do1_client.is_admin}")  # Should be False

# Connect to second data owner
do2_email = "[email protected]"
do2_client = sy.init_session(
    host=do2_email,
    email=ds_email,
    start_syft_event_server=False,
)

print(f"Connected to {do2_email}")

# Get DO1's datasets
do1_datasets = do1_client.dataset.get_all()
for ds in do1_datasets:
    print(f"\nDataset: {ds.name}")
    print(f"  Description: {ds.description}")
    
    # You can access mock data
    mock_path = ds.get_mock_path()
    print(f"  Mock data: {mock_path}")
    
    # But NOT private data (will raise error)
    try:
        private_path = ds.get_private_path()
    except Exception as e:
        print(f"  Private data: ❌ Access denied")

# On first machine
syftbox client
# Enter: [email protected]

# On second machine  
syftbox client
# Enter: [email protected]

# On third machine
syftbox client
# Enter: [email protected]

import os
from pathlib import Path

# Point to your SyftBox config
ds = sy.init_session(host="[email protected]", email="[email protected]")
os.environ["SYFTBOX_CLIENT_CONFIG_PATH"] = str(ds.syftbox_client.config_path)

# Configure logging
os.environ["LOGURU_LEVEL"] = "DEBUG"

# Set message timeout (in seconds)
os.environ["SYFT_FLWR_MSG_TIMEOUT"] = "30"

print("✅ Environment configured")

import syft_flwr
from pathlib import Path

# Your FL project directory
PROJECT_PATH = Path("./fl-diabetes-prediction")

# Mock dataset paths from data owners
mock_paths = [
    do1_client.dataset.get(name="pima-indians-diabetes-database").get_mock_path(),
    do2_client.dataset.get(name="pima-indians-diabetes-database").get_mock_path(),
]

print(f"Mock paths: {mock_paths}")

# Run simulation
syft_flwr.run(PROJECT_PATH, mock_paths)

simulation_logs/
├── client_0.log
├── client_1.log
└── server.log

import os

# Encryption is enabled by default
# To explicitly control:
os.environ["SYFT_FLWR_ENCRYPTION_ENABLED"] = "true"  # Enable
# os.environ["SYFT_FLWR_ENCRYPTION_ENABLED"] = "false"  # Disable (dev only)

from syft_crypto.x3dh_bootstrap import ensure_bootstrap

# This runs automatically in syft_flwr
client = ensure_bootstrap(syftbox_client)
print("🔐 End-to-end encryption enabled")

import syft_flwr

PROJECT_PATH = Path("./fl-diabetes-prediction")
do_emails = ["[email protected]", "[email protected]"]
ds_email = "[email protected]"

syft_flwr.bootstrap(
    PROJECT_PATH,
    aggregator=ds_email,
    datasites=do_emails,
    transport="syftbox",  # Use local SyftBox
)

# Clean before submitting
!rm -rf {PROJECT_PATH / "fl_diabetes_prediction" / "__pycache__"}
!rm -rf {PROJECT_PATH / "simulation_logs"}

# Submit to DO1
do1_client.job.submit(
    name="fl-diabetes-prediction",
    user_code_path=PROJECT_PATH,
    dataset_name="pima-indians-diabetes-database",
    entrypoint="main.py",
)

# Submit to DO2
do2_client.job.submit(
    name="fl-diabetes-prediction",
    user_code_path=PROJECT_PATH,
    dataset_name="pima-indians-diabetes-database",
    entrypoint="main.py",
)

print("✅ Jobs submitted, waiting for approval...")

# Data owner reviews and approves
jobs = do1_client.job.get_all()
pending_job = jobs[0]

print(f"Job from: {pending_job.created_by}")
print(f"Code path: {pending_job.user_code_path}")

# Approve the job
do1_client.job.approve(pending_job)
print("✅ Job approved")

# Data scientist submits server job to themselves
server_job = ds.job.submit(
    name="fl-diabetes-prediction-server",
    user_code_path=PROJECT_PATH,
    entrypoint="main.py",
)

# Auto-approve own job
ds.job.approve(server_job)

# Run the server (blocking)
ds.run_private(server_job, blocking=True)

# View logs
ds.job.show_logs(server_job)

# Check output directory
output_dir = ds.job.get_output_dir(server_job)
print(f"Results saved to: {output_dir}")

# List model checkpoints
weights_dir = output_dir / "weights"
for weight_file in weights_dir.glob("*.safetensors"):
    print(f"  - {weight_file.name}")

# View sync status
syftbox status

# Check recent sync activity
tail -f ~/SyftBox/.syftbox/logs/sync.log

# Remove temporary files
!rm -rf {PROJECT_PATH / "main.py"}
!rm -rf {PROJECT_PATH / "**/__pycache__"}
!rm -rf {PROJECT_PATH / "simulation_logs"}

print("✅ Cleanup complete")