lace-client 0.5.10

Lace: AI Training Transparency Protocol - Prevent copyright lawsuits with cryptographic proof

Lace - AI Training Transparency Protocol

Prevent copyright lawsuits by proving what you DIDN'T train on.

Lace provides cryptographic proof of AI training provenance through loss trajectory monitoring. When model outputs resemble copyrighted content, you can prove definitively whether that content was in your training data.

🚀 Quick Start

pip install lace-client

import lace

# Before training: Create attestation of your dataset
attestation_id = lace.attest("./training_data")

# During training: One-line integration (zero overhead)
lace.monitor()  # Automatically hooks into PyTorch/TensorFlow

# After training: Verify training relationship
result = lace.verify(attestation_id)
print(f"Correlation: {result['correlation']['score']:.3f}")
print(f"Legal verdict: {result['correlation']['verdict']}")

🔑 Get Your API Key

All processing happens in our secure cloud infrastructure for IP protection.

Get your free API key: https://withlace.ai/request-demo

export LACE_API_KEY=your_api_key_here

💡 How It Works

1. Attestation: Before training, Lace creates a cryptographic fingerprint of your dataset
2. Monitoring: During training, Lace captures loss trajectories with zero overhead
3. Correlation: After training, Lace proves the training relationship through loss curve analysis
4. Legal Evidence: Get legally-sufficient evidence for copyright defense

📊 Integration Examples

HuggingFace Transformers

from transformers import Trainer
import lace

# Create attestation
attestation_id = lace.attest("./data")

# Train with monitoring
trainer = Trainer(model, args, dataset)
monitor = lace.monitor(attestation_id)
trainer.train()

# Get correlation
result = monitor.finalize()

PyTorch

import torch
import lace

# Start monitoring
lace.monitor()

# Your normal training loop
for epoch in range(epochs):
    for batch in dataloader:
        loss = model(batch)
        loss.backward()  # Automatically captured!
        optimizer.step()

TensorFlow/Keras

import tensorflow as tf
import lace

# Start monitoring
lace.monitor()

# Your normal training
model.fit(x_train, y_train, epochs=10)  # Automatically captured!

🛡️ Legal Protection

Lace combines multiple verification methods to provide legally defensible evidence:

• Cryptographic proofs that cannot be forged
• Loss trajectory analysis unique to your dataset
• Behavioral verification across multiple metrics
• Bloom filter checks with 99.99% accuracy

The probability of our verification being wrong is negligible - comparable to DNA evidence in court. When accused of training on copyrighted content, you have definitive proof of what was and wasn't in your dataset.

🏢 Enterprise Features

• Unlimited attestations: No limits on dataset size
• Priority support: Direct email support with SLA
• SLA guarantees: 99.9% uptime commitment
• Custom deployment: On-premise options available

Contact: [e-mail]

📖 Documentation

• Docs: https://withlace.ai/docs
• Website: https://withlace.ai
• Examples: Coming soon

🤝 Support

• Email: [e-mail]
• Website: https://withlace.ai

📄 License

Apache 2.0 - See LICENSE for details.

---

Stop worrying about copyright lawsuits. Start building with confidence.

Get Started Free →