haive.core.engine.agent

Agent engine module for the Haive framework.

This module provides the core architecture for all agent implementations in Haive. It delivers consistent schema handling, execution flows, persistence management, and extensibility through patterns.

The agent engine implements protocol-based interfaces to ensure all agent implementations conform to consistent APIs, while providing flexibility for custom agent behaviors and execution patterns.

Key Components:

Agent: Base agent class with graph-based execution AgentConfig: Comprehensive configuration for agents AgentProtocol: Protocol interface for agent implementations PersistenceManager: State persistence and checkpointing Pattern: Reusable agent implementation patterns

The agent engine integrates tightly with LangGraph for execution flow management and provides built-in support for: - Dynamic state schema management - Streaming outputs - State persistence - Graph visualization - Tool integration - Custom execution patterns

Examples

Basic agent creation:

from haive.core.engine.agent import Agent, AgentConfig
from haive.core.engine.aug_llm import AugLLMConfig

agent = Agent(
    name="assistant",
    engine=AugLLMConfig(model="gpt-4")
)

result = agent.invoke("Hello!")

Agent with persistence:

from haive.core.engine.agent import Agent, AgentConfig
from haive.core.engine.agent.persistence import CheckpointerConfig

config = AgentConfig(
    name="persistent_agent",
    engine=AugLLMConfig(model="gpt-4"),
    checkpointer_config=CheckpointerConfig(
        type="sqlite",
        connection_string="agent_state.db"
    )
)

agent = Agent(config)

See also

  • Agent Protocol documentation: protocols.py

  • Persistence system: persistence/

  • Agent patterns: pattern.py

  • Configuration guide: config.py

Submodules