PySyft

PySyft v2

syft 0.10+ is the successor of syft-client. syft is the sync engine (import syft as sy); datasets and jobs live in syft-rds (from syft_rds import login_do, login_ds). If you depend on the legacy PySyft ≤0.9 API, pin syft<0.10.

PySyft allows data scientists run computations which are reviewed and run by data owners on private data. Users securely communicate through cloud storage their organizations already use (Google Drive, Microsoft 365, etc.) or serverless p2p chnannels like webRTC. No new infrastructure required and all communication between be peers can be signed and encrypted.

Docs

• Workflow — End-to-end privacy-preserving data analysis workflow
• API Reference — All public client methods and properties
• Authentication & Setup — Google Cloud OAuth setup for local/Jupyter usage
• Background Services — Email notifications, running api calls, and TUI dashboard
• Connections — How the Google Drive transport layer works
• Permissions - Permissions for syft
• Enclaves - Enclaves with syft

Tutorials

• Double blind LLM evals in enclaves — a researcher evaluates a private model on a private benchmark inside an enclave, with neither side revealing its assets
• LLM user logs analysis — submit jobs against private LLM user logs and get back only the approved results (data owner notebook)

Features

• Privacy-preserving — Private data never leaves the data owner's machine; only approved results are shared
• Transport-agnostic — Works over Google Drive today, extensible to any file-based transport
• Offline-first — Full functionality even when peers are offline; changes sync when connectivity resumes
• Peer-to-peer with explicit auth — Data owners must approve each collaborator before any data flows
• Approved job execution — Jobs only run after approval from the data owner(s), the approval includes access to specific private data
• Dataset sharing with mock/private separation — Data scientists explore mock data, then submit jobs that run on the real thing

Quick Start

Before you start: this quick start assumes Google Colab, where login is a browser pop-up and works out of the box. On a local machine or in Jupyter you first need a Google Cloud OAuth token: follow docs/auth.md (about 15 minutes the first time), then pass it as token_path to login_do / login_ds. Each role needs its own Google account.

We assume two parties here, a Data Owner (DO) and a Data Scientist (DS), the DS wants to do an analysis on private data of the DO. For brevity we use code blocks, but in practice these would be distributed: each party executes their code on their own machine.

# Data scientist
uv pip install "syft>=0.10.0" "syft-rds>=0.6.0"
# Data owner (adds the background services)
uv pip install "syft>=0.10.0" "syft-rds>=0.6.0" "syft-bg>=0.3.12"

import syft as sy                          # sync engine + helpers (resolve_dataset_file_path, bug_report, ...)
from syft_rds import login_do, login_ds    # datasets + jobs (the Remote Data Science client)

# Colab: login is a browser pop-up, nothing to set up
# Local / Jupyter: create a token first, see docs/auth.md
do = login_do(email="[email protected]")  # local: add token_path="token_do.json"
ds = login_ds(email="[email protected]")  # second account; local: token_path="token_ds.json"

# Peer request & approve
ds.add_peer("[email protected]")
do.approve_peer_request("[email protected]")

# Create & sync dataset
do.create_dataset(
    name="census",
    mock_path="mock.txt", # create this and add some text
    private_path="private.txt", # create this and add some text
    users=["[email protected]"],
)
do.sync(); ds.sync()
datasets = ds.datasets.get_all()

Write an analysis.py that reads the dataset and produces a result in our case this is just the length of the data. Inside a job, resolve_dataset_file_path automatically resolves to the private data:

# analysis.py
import json
import syft as sy

data_path = sy.resolve_dataset_file_path("census")
with open(data_path, "r") as f:
    data = f.read()

with open("outputs/result.json", "w") as f:
    json.dump({"length": len(data)}, f)

Submit the job and retrieve results:

# Submit job
ds.submit_python_job(
    user="[email protected]",
    code_path="analysis.py",
    job_name="analysis",
)
ds.sync(); do.sync()

# Data owner Approves & runs job
do.jobs["[email protected]"]["analysis"].approve()
do.process_approved_jobs(share_outputs_with_submitter=True)
do.sync(); ds.sync()
result = open(ds.jobs["[email protected]"]["analysis"].output_paths[0]).read()

Packages

Package	Description
syft-rds	Remote Data Science client: login_do/login_ds, datasets, jobs
syft-datasets	Dataset management and sharing
syft-job	Job submission and execution
syft-permissions	Permission system for Syft datasites
syft-perms	User-facing permission API for Syft datasites
syft-bg	Background services TUI dashboard for SyftBox
syft-notebook-ui	Jupyter notebook display utilities

Development

# Install in development mode
uv pip install -e .

# Run tests
just test-unit          # Unit tests (fast, mocked)
just test-integration   # Integration tests (slow, real API)

---

Built by OpenMined — building open-source technology for privacy-preserving data science and AI.

Support

For questions about PySyft, reach out via #support on Slack.

Community

Supported by the OpenMined Foundation, the OpenMined Community is an online network of over 17,000 technologists, researchers, and industry professionals keen to unlock 1000x more data in every scientific field and industry.

Contributors

OpenMined and Syft appreciates all contributors, if you would like to fix a bug or suggest a new feature, please reach out via Github or Slack!

About OpenMined

OpenMined is a non-profit foundation creating technology infrastructure that helps researchers get answers from data without needing a copy or direct access. Our community of technologists is building Syft.

Supporters

License

Apache License 2.0 Person icons created by Freepik - Flaticon