Agent Persistence with Sherpa

10. Agent Persistence with Sherpa#

Agent persistence allows you to save and restore agents across sessions, maintaining their state and configuration. Sherpa AI provides a comprehensive persistence system with multi-user support and flexible storage backends.

10.1. Basic Agent Persistence#

The PersistentAgentPool allows you to save agents to storage and restore them later, maintaining their configuration and state. This is essential for applications that need to persist agent instances across application restarts or share agents between different sessions.

Let’s create and save an agent:

from sherpa_ai.agents.persistent_agent_pool import PersistentAgentPool
from sherpa_ai.agents.qa_agent import QAAgent

# Create persistent pool
pool = PersistentAgentPool("my_agents.db", "sqlite")

# Create and save an agent
agent = QAAgent(name="My Assistant", description="A helpful QA agent")
agent_id = pool.save_agent(agent, user_id="user123")

# Load agent later
loaded_agent = pool.get_agent(agent_id)
print(f"Loaded: {loaded_agent.name}")

10.2. User Management#

The UserAgentManager provides multi-user support with isolated agent spaces, user preferences, and session tracking. This enables applications to support multiple users where each user can have their own set of agents with personalized settings and usage limits.

For multi-user applications, use the UserAgentManager:

from sherpa_ai.agents.user_agent_manager import UserAgentManager

# Create user manager
manager = UserAgentManager("user_agents.db")

# Create user and agent
manager.create_user("user123", max_agents=5)
agent_id = manager.create_agent_for_user("user123", "My Assistant", "QAAgent")

# Start session and use agent
session_id = manager.start_user_session("user123", "My Assistant")
agent = manager.get_agent_for_user("user123", "My Assistant")
result = agent.synthesize_output()
manager.end_user_session(session_id)

10.3. Storage Backends#

Sherpa supports two storage backends for agent persistence, each optimized for different use cases. SQLite provides robust database functionality with ACID compliance and concurrent access, while JSON offers simplicity for development and single-user scenarios.

SQLite (Recommended for production):

pool = PersistentAgentPool("agents.db", "sqlite")

JSON (Good for development):

pool = PersistentAgentPool("agents.json", "json")

10.4. Setup and Configuration#

Choose your persistence setup based on your application needs. For simple single-user applications, use the basic PersistentAgentPool. For multi-user applications with user management, use the UserAgentManager which provides additional features like user preferences and session tracking.

Basic Setup (Single User):

from sherpa_ai.agents.persistent_agent_pool import PersistentAgentPool

# Choose storage backend
pool = PersistentAgentPool("my_agents.db", "sqlite")  # or "json"

# Save and load agents
agent_id = pool.save_agent(agent, user_id="default")
loaded_agent = pool.get_agent(agent_id)

Multi-User Setup:

from sherpa_ai.agents.user_agent_manager import UserAgentManager

# Initialize with database
manager = UserAgentManager("user_agents.db")

# Create users with custom preferences
manager.create_user("user1", max_agents=10, auto_save=True)
manager.create_user("user2", max_agents=5, default_agent_type="QAAgent")

# Create agents for specific users
agent_id = manager.create_agent_for_user("user1", "Assistant", "QAAgent")

The persistence system supports multi-user isolation, session tracking, and flexible storage options. For detailed API reference, see API Documentation.