Agent_System.md

graph LR
    Agent_LiteAgent["Agent/LiteAgent"]
    ReasoningHandler["ReasoningHandler"]
    CrewAgentExecutor["CrewAgentExecutor"]
    CrewAgentParser["CrewAgentParser"]
    LLM["LLM"]
    ToolUsage["ToolUsage"]
    Memory_Components["Memory Components"]
    Agent_Utilities["Agent Utilities"]
    Agent_LiteAgent -- "delegates execution to" --> CrewAgentExecutor
    Agent_LiteAgent -- "delegates reasoning to" --> ReasoningHandler
    Agent_LiteAgent -- "manages context with" --> Memory_Components
    Agent_LiteAgent -- "utilizes" --> Agent_Utilities
    ReasoningHandler -- "communicates with" --> LLM
    CrewAgentExecutor -- "manages execution for" --> Agent_LiteAgent
    CrewAgentExecutor -- "communicates with" --> LLM
    CrewAgentExecutor -- "passes output to" --> CrewAgentParser
    CrewAgentExecutor -- "invokes" --> ToolUsage
    CrewAgentExecutor -- "accesses/stores context with" --> Memory_Components
    CrewAgentExecutor -- "uses" --> Agent_Utilities
    CrewAgentParser -- "interprets output from" --> LLM
    CrewAgentParser -- "provides interpretation to" --> CrewAgentExecutor
    LLM -- "provides intelligence to" --> CrewAgentExecutor
    LLM -- "provides intelligence to" --> ReasoningHandler
    ToolUsage -- "executes tools for" --> CrewAgentExecutor
    Memory_Components -- "manages context for" --> Agent_LiteAgent
    Memory_Components -- "supplies context to" --> CrewAgentExecutor
    Agent_Utilities -- "supports" --> Agent_LiteAgent
    Agent_Utilities -- "supports" --> CrewAgentExecutor

Loading

Details

The Agent System subsystem encapsulates the intelligence, roles, and decision-making logic of individual AI agents within a crew. It is central to the AI Agent Orchestration Framework's agent-centric design.

Agent/LiteAgent

Initializes and manages the agent's lifecycle, including setting up the executor, formatting messages, and initiating the core execution loop. It serves as the primary entry point for an agent's operations.

Related Classes/Methods:

• src.crewai.lite_agent.py

ReasoningHandler

Guides the agent's thought process by creating and refining plans, constructing prompts, and parsing structured reasoning responses from the LLM. It is critical for the agent's intelligent decision-making.

Related Classes/Methods:

• src/crewai/utilities/reasoning_handler.py

CrewAgentExecutor

Orchestrates the iterative execution flow of agent actions, managing the invocation loop, human feedback, and processing individual agent steps. It is the engine driving the agent's actions.

Related Classes/Methods:

• src/crewai/agents/crew_agent_executor.py

CrewAgentParser

Interprets raw text output from the LLM, extracting structured AgentAction (tool calls) or AgentFinish (final answer) objects. This component is vital for the agent to understand and act upon the LLM's responses.

Related Classes/Methods:

• src/crewai/agents/parser.py

LLM

Acts as a unified interface for various LLM providers, handling API calls, preparing completion parameters, and processing streaming/non-streaming responses, including tool calls. It is the core intelligence provider for the agent.

Related Classes/Methods:

• src/crewai/llm.py

ToolUsage

Facilitates the agent's use of external tools by parsing tool calling formats, selecting, executing, and managing errors or usage limits. This enables the agent to interact with the external environment.

Related Classes/Methods:

• src/crewai/tools/tool_usage.py

Memory Components

Manages different types of agent memory (e.g., short-term, long-term, external, contextual) by storing information and retrieving relevant context for tasks. It provides the agent with state and historical context.

Related Classes/Methods:

• src/crewai/memory

Agent Utilities

Provides common utility functions supporting agent operations, such as handling maximum iteration limits, processing LLM responses, and managing context length. These are supporting functions that ensure the smooth operation of the agent.

Related Classes/Methods:

• src/crewai/utilities/agent_utils.py

FAQ