Tool Use & Function Calling

function executeParallelTools(userRequest, tools):
    response = llm.generate(
        messages: [userRequest],
        tools: tools,
        parallelToolCalls: true
    )

    if response.hasMultipleToolCalls:
        # Execute all tool calls concurrently
        results = parallel.map(response.toolCalls, (call) =>
            tools[call.name].execute(call.arguments)
        )

        # Collect all results
        toolResults = zip(response.toolCalls, results)

        return llm.generate(
            messages: [userRequest, response, ...toolResults]
        )

    return response.content

import asyncio
from langchain_openai import ChatOpenAI
from langchain_core.tools import tool
from langchain_core.messages import HumanMessage, ToolMessage

@tool
async def get_weather(location: str) -> str:
    """Get current weather for a location."""
    return await weather_api.get_async(location)

@tool
async def search_web(query: str) -> str:
    """Search the web for information."""
    return await search_api.search_async(query)

@tool
async def get_stock_price(symbol: str) -> str:
    """Get current stock price."""
    return await stock_api.get_price_async(symbol)

# Bind multiple tools
llm = ChatOpenAI(model="gpt-5.1")
tools = [get_weather, search_web, get_stock_price]
llm_with_tools = llm.bind_tools(tools)

async def agent_with_parallel_tools(query: str) -> str:
    messages = [HumanMessage(content=query)]

    response = await llm_with_tools.ainvoke(messages)

    if response.tool_calls:
        messages.append(response)

        # Execute all tools in parallel
        tool_tasks = []
        for tool_call in response.tool_calls:
            tool_fn = next(t for t in tools if t.name == tool_call["name"])
            tool_tasks.append(tool_fn.ainvoke(tool_call["args"]))

        results = await asyncio.gather(*tool_tasks)

        # Add all results
        for tool_call, result in zip(response.tool_calls, results):
            messages.append(
                ToolMessage(content=result, tool_call_id=tool_call["id"])
            )

        return (await llm_with_tools.ainvoke(messages)).content

    return response.content

using Microsoft.Agents.AI;
using Microsoft.Extensions.AI;
using System.ComponentModel;

// Define multiple tools as functions
[Description("Get weather for a location")]
static async Task<string> GetWeather(string location)
    => await weatherService.GetAsync(location);

[Description("Search the web")]
static async Task<string> SearchWeb(string query)
    => await searchService.SearchAsync(query);

[Description("Get stock price")]
static async Task<string> GetStockPrice(string symbol)
    => await stockService.GetPriceAsync(symbol);

// Create agent with multiple tools
AIAgent agent = new AzureOpenAIClient(endpoint, credentials)
    .GetChatClient("gpt-5.1")
    .AsAIAgent(
        instructions: "You are a helpful assistant",
        tools: [
            AIFunctionFactory.Create(GetWeather),
            AIFunctionFactory.Create(SearchWeb),
            AIFunctionFactory.Create(GetStockPrice)
        ]
    );

// Agent automatically handles parallel tool execution
var result = await agent.RunAsync(
    "What's the weather in NYC and Tokyo, and Apple's stock price?"
);

function executeWithRetry(toolCall, maxRetries = 3):
    for attempt in range(maxRetries):
        try:
            result = tools[toolCall.name].execute(toolCall.args)
            return { success: true, data: result }
        catch error:
            if isRetryable(error) and attempt < maxRetries - 1:
                wait(exponentialBackoff(attempt))
                continue
            else:
                return {
                    success: false,
                    error: formatError(error)
                }

function agentLoop(userRequest):
    while not complete:
        response = llm.generate(messages, tools)

        if response.hasToolCall:
            result = executeWithRetry(response.toolCall)

            if not result.success:
                # Let the model know about the failure
                messages.append(toolError(result.error))
                # Model can try a different approach
            else:
                messages.append(toolResult(result.data))
        else:
            return response.content

from langchain_core.tools import tool, ToolException
from langchain_core.runnables import RunnableConfig
from tenacity import retry, stop_after_attempt, wait_exponential

# Tools can handle their own errors gracefully
@tool
def search_database(query: str) -> str:
    """Search the database with automatic error handling."""
    try:
        return db.search(query)
    except DatabaseError as e:
        # Return error message instead of raising
        raise ToolException(f"Database error: {e}")

# Convert ToolException into an error message for the model
search_database.handle_tool_error = True

# Custom retry wrapper for tools
class RetryableTool:
    def __init__(self, tool_fn, max_retries=3):
        self.tool = tool_fn
        self.max_retries = max_retries

    @retry(
        stop=stop_after_attempt(3),
        wait=wait_exponential(multiplier=1, min=1, max=10)
    )
    def invoke(self, args: dict) -> str:
        return self.tool.invoke(args)

    def safe_invoke(self, args: dict) -> dict:
        try:
            result = self.invoke(args)
            return {"success": True, "data": result}
        except Exception as e:
            return {
                "success": False,
                "error": str(e),
                "error_type": type(e).__name__
            }

# Using LangGraph for agent with error recovery
from langgraph.prebuilt import create_react_agent, ToolNode

# ToolNode catches tool exceptions and returns them as
# ToolMessages, so the agent can see errors and adapt
agent = create_react_agent(
    llm,
    ToolNode(tools, handle_tool_errors=True)
)

# The agent will receive error messages and can adapt
result = agent.invoke({"messages": [("user", query)]})

using Microsoft.Agents.AI;
using Microsoft.Extensions.AI;
using System.ComponentModel;
using Polly;

// Create retry policy
var retryPolicy = Policy
    .Handle<HttpRequestException>()
    .Or<TimeoutException>()
    .WaitAndRetryAsync(
        3,
        attempt => TimeSpan.FromSeconds(Math.Pow(2, attempt)),
        onRetry: (ex, time, attempt, ctx) =>
            Console.WriteLine($"Retry {attempt}: {ex.Message}")
    );

// Define tool with built-in error handling
[Description("Search with automatic retry")]
async Task<string> SearchWithRetry(string query)
{
    try
    {
        return await retryPolicy.ExecuteAsync(async () =>
        {
            return await searchService.SearchAsync(query);
        });
    }
    catch (Exception ex)
    {
        // Return error info so agent can adapt
        return $"Error: {ex.Message}. Try a different approach.";
    }
}

// Create agent with resilient tools
AIAgent agent = new AzureOpenAIClient(endpoint, credentials)
    .GetChatClient("gpt-5.1")
    .AsAIAgent(
        instructions: "You are a helpful assistant",
        tools: [AIFunctionFactory.Create(SearchWithRetry)]
    );

// Agent receives error messages and can adapt its approach
var result = await agent.RunAsync(prompt);