Implementation:Microsoft Autogen AssistantAgent Init Tools

Knowledge Sources	Microsoft AutoGen
Domains	Tool Use, LLM Agents, Function Calling, Agent Configuration
Last Updated	2026-02-11 00:00 GMT

Overview

Concrete tool for creating LLM agents with tool-use capabilities provided by Microsoft AutoGen, focusing on the tool-related configuration parameters of AssistantAgent.__init__.

Description

The AssistantAgent constructor accepts tool-related parameters that configure how the agent discovers, invokes, and processes tool results. Tools can be provided directly as a list of BaseTool instances or plain callables (which are auto-wrapped into FunctionTool using their docstrings as descriptions). Alternatively, tools can be provided via one or more Workbench instances that manage tool lifecycle and discovery.

The tools and workbench parameters are mutually exclusive. When tools is provided, the agent internally wraps them in a StaticStreamWorkbench. When workbench is provided (a single workbench or a sequence), the agent delegates all tool management to those workbenches.

The constructor validates that the model client supports function calling before accepting tools. It also validates that all tool names are unique and do not conflict with handoff tool names.

The reflect_on_tool_use parameter controls whether the agent makes an additional LLM call after tool execution to interpret the results. The max_tool_iterations parameter controls how many rounds of tool calls the agent can perform. The tool_call_summary_format and tool_call_summary_formatter parameters control how tool results are formatted when reflection is disabled.

Usage

Use these parameters when creating an AssistantAgent that needs tool-use capabilities. Pass tools for simple setups with known tool lists, or pass workbench for advanced scenarios involving MCP servers or custom tool management. Configure reflect_on_tool_use and max_tool_iterations to control the agent's tool-use behavior.

Code Reference

Source Location

Repository: Microsoft AutoGen
File: python/packages/autogen-agentchat/src/autogen_agentchat/agents/_assistant_agent.py (lines 724-859)

Signature

class AssistantAgent:
    def __init__(
        self,
        name: str,
        model_client: ChatCompletionClient,
        *,
        tools: List[BaseTool[Any, Any] | Callable[..., Any] | Callable[..., Awaitable[Any]]] | None = None,
        workbench: Workbench | Sequence[Workbench] | None = None,
        handoffs: List[HandoffBase | str] | None = None,
        model_context: ChatCompletionContext | None = None,
        description: str = "An agent that provides assistance with ability to use tools.",
        system_message: str | None = "You are a helpful AI assistant. Solve tasks using your tools. Reply with TERMINATE when the task has been completed.",
        model_client_stream: bool = False,
        reflect_on_tool_use: bool | None = None,
        max_tool_iterations: int = 1,
        tool_call_summary_format: str = "{result}",
        tool_call_summary_formatter: Callable[[FunctionCall, FunctionExecutionResult], str] | None = None,
        output_content_type: type[BaseModel] | None = None,
        output_content_type_format: str | None = None,
        memory: Sequence[Memory] | None = None,
        metadata: Dict[str, str] | None = None,
    ):
        ...

Import

from autogen_agentchat.agents import AssistantAgent

I/O Contract

Inputs (Tool-Related Parameters)

Name	Type	Required	Description
tools	Callable \| AsyncCallable] or None	No	A list of tools for the agent. Each element can be a `BaseTool` instance (e.g., `FunctionTool`) or a plain callable. Callables are auto-wrapped into `FunctionTool` using their docstring as the description. Cannot be used together with `workbench`.
workbench	Workbench or Sequence[Workbench] or None	No	One or more workbench instances that manage tool lifecycle and discovery. Cannot be used together with `tools`. A single workbench is wrapped in a list internally.
reflect_on_tool_use	bool or None	No	Whether to make an additional LLM call after tool execution to interpret results. Defaults to None, which resolves to True if `output_content_type` is set, otherwise False.
max_tool_iterations	int	No	Maximum number of tool call loop iterations. Must be >= 1. Defaults to 1. Higher values allow the agent to call tools multiple times before producing a final response.
tool_call_summary_format	str	No	Format string for summarizing tool results when reflection is disabled. Uses `{result}` as placeholder. Defaults to `"{result}"`.
tool_call_summary_formatter	Callable or None	No	A custom function for formatting tool call summaries. Takes a `FunctionCall` and `FunctionExecutionResult` and returns a string. Overrides `tool_call_summary_format` when provided.

Outputs

Name	Type	Description
instance	AssistantAgent	A configured agent instance with tool-use capabilities. The agent can be used in teams or run directly via `run()` or `run_stream()`.

Usage Examples

Basic Example with Tools List

from autogen_agentchat.agents import AssistantAgent
from autogen_ext.models.openai import OpenAIChatCompletionClient
from autogen_core.tools import FunctionTool


async def search_web(query: str) -> str:
    """Search the web for information."""
    return f"Results for: {query}"


model_client = OpenAIChatCompletionClient(model="gpt-4o")

agent = AssistantAgent(
    name="search_agent",
    model_client=model_client,
    tools=[search_web],  # Auto-wrapped using docstring as description
)

Multi-Iteration Tool Use with Reflection

from autogen_agentchat.agents import AssistantAgent
from autogen_ext.models.openai import OpenAIChatCompletionClient
from autogen_core.tools import FunctionTool


async def lookup_database(query: str) -> str:
    """Query the internal database."""
    return f"DB result: {query}"


async def verify_fact(claim: str) -> str:
    """Verify a factual claim."""
    return f"Verified: {claim} is correct"


model_client = OpenAIChatCompletionClient(model="gpt-4o")

agent = AssistantAgent(
    name="research_agent",
    model_client=model_client,
    tools=[lookup_database, verify_fact],
    reflect_on_tool_use=True,
    max_tool_iterations=3,
)

Using a Workbench

from autogen_agentchat.agents import AssistantAgent
from autogen_ext.models.openai import OpenAIChatCompletionClient
from autogen_ext.tools.mcp import McpWorkbench, StdioServerParams


server_params = StdioServerParams(
    command="npx",
    args=["-y", "@modelcontextprotocol/server-filesystem", "/tmp"],
)

workbench = McpWorkbench(server_params=server_params)

model_client = OpenAIChatCompletionClient(model="gpt-4o")

agent = AssistantAgent(
    name="file_agent",
    model_client=model_client,
    workbench=workbench,
)

Related Pages

Implements Principle

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment