Memory Shield

Tamper-proof conversation history

Memory Shield protects AI agent memory from manipulation attacks.

Using HMAC-based integrity verification, it detects any unauthorized modifications

to conversation history, preventing context poisoning and memory injection attacks.

python

from sentinelseed import MemoryShield

# Initialize with secret key
shield = MemoryShield(secret_key="your-secret-key")

# Add messages with integrity
shield.add_message("user", "Transfer $100 to account X")
shield.add_message("assistant", "Processing transfer...")

# Verify integrity before processing
if shield.verify_integrity():
    # Safe to proceed
    response = agent.process(shield.get_messages())
else:
    # Tampering detected!
    shield.rollback_to_last_valid()

Key Features

HMAC-SHA256 integrity signing

Per-message tamper detection

Automatic rollback on corruption

Zero-knowledge verification

Multi-turn conversation protection

Best For

Long-running autonomous agents

Multi-agent communication

Sensitive conversation handling

Audit trail requirements

Database Guard