agent

Functions

create_default_planner(tool_manager[, ...])	Create a default planner with the given model client, model kwargs, template, task desc, cache path, use cache, max steps.
create_default_tool_manager([tools, ...])	Create a default tool manager with the given tools, context variables, and add_llm_as_fallback.

Classes

Agent(name, tools, context_variables, ...)

A high-level agentic component that orchestrates AI planning and tool execution.

class Agent(name: str, tools: ~typing.List[~typing.Any] | None = None, context_variables: ~typing.Dict | None = None, add_llm_as_fallback: bool | None = False, model_client: ~adalflow.core.model_client.ModelClient | None = None, model_kwargs: ~typing.Dict[str, ~typing.Any] | None = {}, model_type: ~adalflow.core.types.ModelType | None = ModelType.LLM, template: str | None = None, role_desc: str | ~adalflow.core.prompt_builder.Prompt | None = None, cache_path: str | None = None, use_cache: bool | None = True, answer_data_type: ~typing.Type[~components.agent.agent.T] | None = <class 'str'>, max_steps: int | None = 10, is_thinking_model: bool | None = False, tool_manager: ~adalflow.core.tool_manager.ToolManager | None = None, planner: ~adalflow.core.generator.Generator | None = None, **kwargs)

Bases: Component

A high-level agentic component that orchestrates AI planning and tool execution.

The Agent combines a Generator-based planner for task decomposition with a ToolManager for function calling. It uses a ReAct (Reasoning and Acting) architecture to iteratively plan steps and execute tools to solve complex tasks.

The Agent comes with default prompt templates for agentic reasoning, automatic tool definition integration, and step history tracking. It includes built-in helper tools: - llm_tool: Fallback tool using LLM world knowledge for simple queries

Architecture:

Agent contains two main components: 1. Planner (Generator): Plans and reasons about next actions using an LLM 2. ToolManager: Manages and executes available tools/functions

name

Unique identifier for the agent instance

Type:

str

tool_manager

Manages available tools and their execution

Type:

ToolManager

planner

LLM-based planner for task decomposition and reasoning

Type:

Generator

answer_data_type

Expected type for the final answer output

Type:

Type

max_steps

Maximum number of planning steps allowed

Type:

int

is_thinking_model

Whether the underlying model supports chain-of-thought

Type:

bool

is_training() → bool

get_prompt(**kwargs) → str

Get formatted prompt using generator’s prompt template.

Parameters:

**kwargs – Additional arguments to pass to the generator’s get_prompt

Returns:

Formatted prompt string