syft-nsai

Bridge Data and AI. Privately.

An OpenAI-compatible interface for using SyftBox datasets with AI models in secure enclaves.

🚀 What is syft-nsai?

syft-nsai transforms how you work with federated datasets by providing a familiar OpenAI-style chat completions API that operates within secure enclaves. No more complex setup or unfamiliar APIs - just select your datasets and chat with your data.

import syft_datasets as syd
import syft_nsai as nsai

# Discover datasets with beautiful interactive UI
syd.datasets

# Use them with familiar OpenAI-style API
selected_datasets = [syd.datasets[i] for i in [0, 1, 5]]
response = nsai.client.chat.completions.create(
    model=selected_datasets,  # Your datasets become the "model"
    messages=[{"role": "user", "content": "What trends do you see in this data?"}]
)

# Access results
print(response.choices[0].message.content)

✨ Key Features

• 🔒 Privacy-First: All processing happens in secure enclaves
• 🤝 OpenAI Compatible: Drop-in replacement for familiar chat completions API
• 📊 Real Data Integration: Datasets are loaded and summarized before AI analysis
• ⚡ Lazy Loading: Results are fetched asynchronously for better UX
• 🔄 Seamless Integration: Works perfectly with syft-datasets for dataset discovery

🏗️ Architecture

┌─────────────────┐    ┌──────────────────┐    ┌─────────────────┐
│  syft-datasets  │    │   syft-nsai      │    │  Secure Enclave │
│                 │    │                  │    │                 │
│ Dataset Discovery│───▶│OpenAI-style API  │───▶│  AI Processing  │
│ Interactive UI   │    │Chat Completions  │    │  with Real Data │
│ Search & Select  │    │Async Results     │    │                 │
└─────────────────┘    └──────────────────┘    └─────────────────┘

📦 Installation

pip install syft-nsai

For development:

pip install syft-nsai[dev]

🔑 Configuration

Tinfoil API Key

syft-nsai uses syft-wallet for secure API key management. When you first use NSAI, it will automatically:

1. Check syft-wallet for your Tinfoil API key
2. Check environment variables as a fallback
3. Prompt you to enter the key if not found anywhere
4. Save it securely to syft-wallet for future use

First-time Setup

import syft_nsai as nsai

# On first use, you'll be prompted for your API key
response = nsai.client.chat.completions.create(
    model=datasets,
    messages=[{"role": "user", "content": "Hello"}]
)
# 🔑 Tinfoil API Key Required
# NSAI needs a Tinfoil API key to run AI models in secure enclaves.
# You can get a free API key at: https://tinfoil.sh/
# Please enter your Tinfoil API key: tk_xxxxx...
# ✓ API key saved securely to syft-wallet

Manual Key Management

import syft_nsai as nsai

# Set or update your API key
nsai.set_tinfoil_api_key("tk_your_new_key_here")
# or prompt for input:
nsai.set_tinfoil_api_key()

# Check current key status
nsai.show_tinfoil_status()
# ✓ Tinfoil API key found: tk_SxHBk...UOFd

# Get the key programmatically
api_key = nsai.get_tinfoil_api_key()

Environment Variable Fallback

You can still use environment variables:

export TINFOIL_API_KEY="tk_your_key_here"

NSAI will detect this and offer to save it to syft-wallet for easier future use.

Security Benefits:

• ✅ Keys stored in 1Password (if available) or secure system keyring
• ✅ No hardcoded keys in your code
• ✅ One-time setup per machine
• ✅ Automatic detection across all methods

🚀 Quick Start

1. Discover Datasets

First, use syft-datasets to explore available datasets:

import syft_datasets as syd

# Show interactive dataset browser
syd.datasets

2. Select and Analyze

Use the selected datasets with the OpenAI-compatible API:

import syft_nsai as nsai

# Select datasets (from the interactive UI or programmatically)
my_datasets = [syd.datasets[0], syd.datasets[3]]

# Create chat completion
response = nsai.client.chat.completions.create(
    model=my_datasets,
    messages=[
        {"role": "system", "content": "You are a data analyst."},
        {"role": "user", "content": "Summarize the key insights from this data."}
    ]
)

# Results are loaded lazily when accessed
insights = response.choices[0].message.content
print(insights)

3. Advanced Usage

# Multiple datasets for cross-analysis
crop_data = syd.datasets.search("crop")[0]
weather_data = syd.datasets.search("weather")[0]

response = nsai.client.chat.completions.create(
    model=[crop_data, weather_data],
    messages=[{
        "role": "user", 
        "content": "Analyze the correlation between weather patterns and crop yields."
    }]
)

🔧 How It Works

1. Dataset Integration: Selected datasets are loaded and analyzed within the enclave
2. Context Enhancement: Your prompts are enhanced with actual dataset summaries and sample data
3. Secure Processing: AI analysis happens in a privacy-preserving enclave using Tinfoil API
4. Familiar Interface: Results are returned through OpenAI-compatible response objects

🔗 Integration with SyftBox Ecosystem

syft-nsai is part of the broader SyftBox ecosystem:

• syft-datasets: Dataset discovery and management
• syft-core: Core SyftBox functionality
• syftbox-enclave: Secure enclave execution
• syft-rds: Remote data science capabilities

📚 API Reference

Client

import syft_nsai as nsai

# Global client instance
client = nsai.client

Chat Completions

response = nsai.client.chat.completions.create(
    model=datasets_list,        # List of Dataset objects
    messages=messages_list,     # OpenAI-style messages
    **kwargs                    # Additional parameters
)

Response Objects

# ChatCompletion
response.choices              # List of Choice objects

# Choice  
choice = response.choices[0]
choice.message                # Message object

# Message
message = choice.message
message.content               # Actual AI response (lazy loaded)

API Key Management

# Set or update Tinfoil API key
nsai.set_tinfoil_api_key(api_key=None)  # Prompts if None

# Get current API key
api_key = nsai.get_tinfoil_api_key()    # Raises ValueError if not found

# Check key status
nsai.show_tinfoil_status()              # Shows masked key or missing status

🧪 Development

Setup

git clone https://github.com/OpenMined/syft-nsai
cd syft-nsai
uv sync

Testing

uv run pytest

Linting

uv run ruff check .
uv run ruff format .

🤝 Contributing

We welcome contributions! Please see CONTRIBUTING.md for guidelines.

📄 License

Licensed under the Apache License, Version 2.0. See LICENSE for details.

🙏 Acknowledgments

Built with ❤️ by the OpenMined community. Special thanks to:

• The Tinfoil team for secure AI infrastructure
• SyftBox contributors for the federated data ecosystem
• The broader privacy-preserving ML community

---

Transform your data science workflow. Privately and securely.

Get Started → | Documentation → | Examples →