agent-memory-hub 0.1.0

Enterprise-grade agent memory management solution with region governance.

Agent Memory Hub

Enterprise-grade agent memory management solution with region governance.

agent-memory-hub provides a standardized, secure, and compliant way for AI agents to store and recall memories. It enforces region residency requirements (data sovereignty) and provides a clean abstraction over storage backends, defaulting to Google Cloud.

Features

• Session-based Memory: Isolate memory by agent and session ID.
• Region Governance: Enforce data residency (e.g., us-central1, europe-west1).
• Backend Agnostic: Adapter pattern supports multiple backends (Default: Google ADK/GCS).
• Enterprise Security: No hardcoded secrets, strictly typed, and compliance-ready.

Installation

pip install agent-memory-hub

Quick Start

from agent_memory_hub import MemoryClient

# Initialize the client with strict region requirements
memory = MemoryClient(
    agent_id="travel_agent",
    session_id="sess_001",
    region="asia-south1",
    region_restricted=True
)

# Store a memory (will fail if backend is not in asia-south1)
memory.write("User prefers vegetarian food", "episodic")

# Recall memory
print(memory.recall("episodic"))

Region Governance

This SDK is designed for global deployments where data sovereignty is critical. When region_restricted=True is set, the SDK performs a handshake with the control plane to verify that the underlying storage is physically located in the requested region before writing any data.

GCP Prerequisites

By default, this package uses Google Cloud Storage as the backing store.

1. Authentication: Ensure GOOGLE_APPLICATION_CREDENTIALS is set or you are authenticated via gcloud auth application-default login.
2. Permissions: The service account requires storage.objects.create and storage.objects.get permissions on the target bucket.
3. Buckets: Buckets should be named following the convention memory-hub-{region}-{environment} (e.g., memory-hub-asia-south1-prod).

Security & Compliance

• No Secrets: This library does not handle secrets directly. Use IAM roles.
• Dependency Pinning: All dependencies are pinned to secure versions.
• Audit: Compatible with pip-audit and bandit.

See SECURITY.md for vulnerability disclosure.