Skip to main content

Build an Agent

By themselves, language models can't take actions - they just output text. A big use case for LangChain is creating agents. Agents are systems that use an LLM as a reasoning enginer to determine which actions to take and what the inputs to those actions should be. The results of those actions can then be fed back into the agent and it determine whether more actions are needed, or whether it is okay to finish.

In this tutorial we will build an agent that can interact with multiple different tools: one being a local database, the other being a search engine. You will be able to ask this agent questions, watch it call tools, and have conversations with it.

Concepts​

Concepts we will cover are:

  • Using language models, in particular their tool calling ability
  • Creating a Retriever to expose specific information to our agent
  • Using a Search Tool to look up things online
  • Using LangGraph Agents which use an LLM to think about what to do and then execute upon that
  • Debugging and tracing your application using LangSmith

Setup​

Jupyter Notebook​

This guide (and most of the other guides in the documentation) uses Jupyter notebooks and assumes the reader is as well. Jupyter notebooks are perfect for learning how to work with LLM systems because oftentimes things can go wrong (unexpected output, API down, etc) and going through guides in an interactive environment is a great way to better understand them.

This and other tutorials are perhaps most conveniently run in a Jupyter notebook. See here for instructions on how to install.

Installation​

To install LangChain run:

pip install langchain

For more details, see our Installation guide.

LangSmith​

Many of the applications you build with LangChain will contain multiple steps with multiple invocations of LLM calls. As these applications get more and more complex, it becomes crucial to be able to inspect what exactly is going on inside your chain or agent. The best way to do this is with LangSmith.

After you sign up at the link above, make sure to set your environment variables to start logging traces:

export LANGCHAIN_TRACING_V2="true"
export LANGCHAIN_API_KEY="..."

Or, if in a notebook, you can set them with:

import getpass
import os

os.environ["LANGCHAIN_TRACING_V2"] = "true"
os.environ["LANGCHAIN_API_KEY"] = getpass.getpass()

Define tools​

We first need to create the tools we want to use. We will use two tools: Tavily (to search online) and then a retriever over a local index we will create

Tavily​

We have a built-in tool in LangChain to easily use Tavily search engine as tool. Note that this requires an API key - they have a free tier, but if you don't have one or don't want to create one, you can always ignore this step.

Once you create your API key, you will need to export that as:

export TAVILY_API_KEY="..."
from langchain_community.tools.tavily_search import TavilySearchResults
search = TavilySearchResults(max_results=2)
search.invoke("what is the weather in SF")
[{'url': 'https://weather.com/weather/tenday/l/San Francisco CA USCA0987:1:US',
  'content': "Comfy & Cozy\nThat's Not What Was Expected\nOutside\n'No-Name Storms' In Florida\nGifts From On High\nWhat To Do For Wheezing\nSurviving The Season\nStay Safe\nAir Quality Index\nAir quality is considered satisfactory, and air pollution poses little or no risk.\n Health & Activities\nSeasonal Allergies and Pollen Count Forecast\nNo pollen detected in your area\nCold & Flu Forecast\nFlu risk is low in your area\nWe recognize our responsibility to use data and technology for good. recents\nSpecialty Forecasts\n10 Day Weather-San Francisco, CA\nToday\nMon 18 | Day\nConsiderable cloudiness. Tue 19\nTue 19 | Day\nLight rain early...then remaining cloudy with showers in the afternoon. Wed 27\nWed 27 | Day\nOvercast with rain showers at times."},
 {'url': 'https://www.accuweather.com/en/us/san-francisco/94103/hourly-weather-forecast/347629',
  'content': 'Hourly weather forecast in San Francisco, CA. Check current conditions in San Francisco, CA with radar, hourly, and more.'}]

Retriever​

We will also create a retriever over some data of our own. For a deeper explanation of each step here, see this tutorial.

from langchain_community.document_loaders import WebBaseLoader
from langchain_community.vectorstores import FAISS
from langchain_openai import OpenAIEmbeddings
from langchain_text_splitters import RecursiveCharacterTextSplitter

loader = WebBaseLoader("https://docs.smith.langchain.com/overview")
docs = loader.load()
documents = RecursiveCharacterTextSplitter(
    chunk_size=1000, chunk_overlap=200
).split_documents(docs)
vector = FAISS.from_documents(documents, OpenAIEmbeddings())
retriever = vector.as_retriever()
retriever.invoke("how to upload a dataset")[0]
Document(page_content='import Clientfrom langsmith.evaluation import evaluateclient = Client()# Define dataset: these are your test casesdataset_name = "Sample Dataset"dataset = client.create_dataset(dataset_name, description="A sample dataset in LangSmith.")client.create_examples(    inputs=[        {"postfix": "to LangSmith"},        {"postfix": "to Evaluations in LangSmith"},    ],    outputs=[        {"output": "Welcome to LangSmith"},        {"output": "Welcome to Evaluations in LangSmith"},    ],    dataset_id=dataset.id,)# Define your evaluatordef exact_match(run, example):    return {"score": run.outputs["output"] == example.outputs["output"]}experiment_results = evaluate(    lambda input: "Welcome " + input[\'postfix\'], # Your AI system goes here    data=dataset_name, # The data to predict and grade over    evaluators=[exact_match], # The evaluators to score the results    experiment_prefix="sample-experiment", # The name of the experiment    metadata={      "version": "1.0.0",      "revision_id":', metadata={'source': 'https://docs.smith.langchain.com/overview', 'title': 'Getting started with LangSmith | 🦜️🛠️ LangSmith', 'description': 'Introduction', 'language': 'en'})

Now that we have populated our index that we will do doing retrieval over, we can easily turn it into a tool (the format needed for an agent to properly use it)

from langchain.tools.retriever import create_retriever_tool
retriever_tool = create_retriever_tool(
    retriever,
    "langsmith_search",
    "Search for information about LangSmith. For any questions about LangSmith, you must use this tool!",
)

Tools​

Now that we have created both, we can create a list of tools that we will use downstream.

tools = [search, retriever_tool]

Using Language Models​

Next, let's learn how to use a language model by to call tools. LangChain supports many different language models that you can use interchangably - select the one you want to use below!

pip install -qU langchain-openai
import getpass
import os

os.environ["OPENAI_API_KEY"] = getpass.getpass()

from langchain_openai import ChatOpenAI

model = ChatOpenAI(model="gpt-4")
# | output: false
# | echo: false

from langchain_openai import ChatOpenAI

model = ChatOpenAI(model="gpt-4")

You can call the language model by passing in a list of messages. By default, the response is a content string.

from langchain_core.messages import HumanMessage

response = model.invoke([HumanMessage(content="hi!")])
response.content
'Hello! How can I assist you today?'

We can now see what it is like to enable this model to do tool calling. In order to enable that we use .bind_tools to give the language model knowledge of these tools

model_with_tools = model.bind_tools(tools)

We can now call the model. Let's first call it with a normal message, and see how it responds. We can look at both the content field as well as the tool_calls field.

response = model_with_tools.invoke([HumanMessage(content="Hi!")])

print(f"ContentString: {response.content}")
print(f"ToolCalls: {response.tool_calls}")
ContentString: Hello! How can I assist you today?
ToolCalls: []

Now, let's try calling it with some input that would expect a tool to be called.

response = model_with_tools.invoke([HumanMessage(content="What's the weather in SF?")])

print(f"ContentString: {response.content}")
print(f"ToolCalls: {response.tool_calls}")
ContentString: 
ToolCalls: [{'name': 'tavily_search_results_json', 'args': {'query': 'current weather in SF'}, 'id': 'call_nfE1XbCqZ8eJsB8rNdn4MQZQ'}]

We can see that there's now no content, but there is a tool call! It wants us to call the Tavily Search tool.

This isn't calling that tool yet - it's just telling us to. In order to actually calll it, we'll want to create our agent.

Create the agent​

Now that we have defined the tools and the LLM, we can create the agent. We will be using LangGraph to construct the agent. Currently we are using a high level interface to construct the agent, but the nice thing about LangGraph is that this high-level interface is backed by a low-level, highly controllable API in case you want to modify the agent logic.

Now, we can initalize the agent with the LLM and the tools.

Note that we are passing in the model, not model_with_tools. That is because create_tool_calling_executor will call .bind_tools for us under the hood.

from langgraph.prebuilt import chat_agent_executor

agent_executor = chat_agent_executor.create_tool_calling_executor(model, tools)

Run the agent​

We can now run the agent on a few queries! Note that for now, these are all stateless queries (it won't remember previous interactions). Note that the agent will return the final state at the end of the interaction (which includes any inputs, we will see later on how to get only the outputs).

First up, let's how it responds when there's no need to call a tool:

response = agent_executor.invoke({"messages": [HumanMessage(content="hi!")]})

response["messages"]
[HumanMessage(content='hi!', id='1535b889-10a5-45d0-a1e1-dd2e60d4bc04'),
 AIMessage(content='Hello! How can I assist you today?', response_metadata={'token_usage': {'completion_tokens': 10, 'prompt_tokens': 129, 'total_tokens': 139}, 'model_name': 'gpt-4', 'system_fingerprint': None, 'finish_reason': 'stop', 'logprobs': None}, id='run-2c94c074-bdc9-4f01-8fd7-71cfc4777d55-0')]

In order to see exactly what is happening under the hood (and to make sure it's not calling a tool) we can take a look at the LangSmith trace

Let's now try it out on an example where it should be invoking the retriever

response = agent_executor.invoke(
    {"messages": [HumanMessage(content="how can langsmith help with testing?")]}
)
response["messages"]
[HumanMessage(content='how can langsmith help with testing?', id='04f4fe8f-391a-427c-88af-1fa064db304c'),
 AIMessage(content='', additional_kwargs={'tool_calls': [{'id': 'call_FNIgdO97wo51sKx3XZOGLHqT', 'function': {'arguments': '{\n  "query": "how can LangSmith help with testing"\n}', 'name': 'langsmith_search'}, 'type': 'function'}]}, response_metadata={'token_usage': {'completion_tokens': 22, 'prompt_tokens': 135, 'total_tokens': 157}, 'model_name': 'gpt-4', 'system_fingerprint': None, 'finish_reason': 'tool_calls', 'logprobs': None}, id='run-51f6ea92-84e1-43a5-b1f2-bc0c12d8613f-0', tool_calls=[{'name': 'langsmith_search', 'args': {'query': 'how can LangSmith help with testing'}, 'id': 'call_FNIgdO97wo51sKx3XZOGLHqT'}]),
 ToolMessage(content="Getting started with LangSmith | 🦜️🛠️ LangSmith\n\nSkip to main contentLangSmith API DocsSearchGo to AppQuick StartUser GuideTracingEvaluationProduction Monitoring & AutomationsPrompt HubProxyPricingSelf-HostingCookbookQuick StartOn this pageGetting started with LangSmithIntroduction\u200bLangSmith is a platform for building production-grade LLM applications. It allows you to closely monitor and evaluate your application, so you can ship quickly and with confidence. Use of LangChain is not necessary - LangSmith works on its own!Install LangSmith\u200bWe offer Python and Typescript SDKs for all your LangSmith needs.PythonTypeScriptpip install -U langsmithyarn add langchain langsmithCreate an API key\u200bTo create an API key head to the setting pages. Then click Create API Key.Setup your environment\u200bShellexport LANGCHAIN_TRACING_V2=trueexport LANGCHAIN_API_KEY=<your-api-key># The below examples use the OpenAI API, though it's not necessary in generalexport OPENAI_API_KEY=<your-openai-api-key>Log your first trace\u200bWe provide multiple ways to log traces\n\nLearn about the workflows LangSmith supports at each stage of the LLM application lifecycle.Pricing: Learn about the pricing model for LangSmith.Self-Hosting: Learn about self-hosting options for LangSmith.Proxy: Learn about the proxy capabilities of LangSmith.Tracing: Learn about the tracing capabilities of LangSmith.Evaluation: Learn about the evaluation capabilities of LangSmith.Prompt Hub Learn about the Prompt Hub, a prompt management tool built into LangSmith.Additional Resources\u200bLangSmith Cookbook: A collection of tutorials and end-to-end walkthroughs using LangSmith.LangChain Python: Docs for the Python LangChain library.LangChain Python API Reference: documentation to review the core APIs of LangChain.LangChain JS: Docs for the TypeScript LangChain libraryDiscord: Join us on our Discord to discuss all things LangChain!FAQ\u200bHow do I migrate projects between organizations?\u200bCurrently we do not support project migration betwen organizations. While you can manually imitate this by\n\nteam deals with sensitive data that cannot be logged. How can I ensure that only my team can access it?\u200bIf you are interested in a private deployment of LangSmith or if you need to self-host, please reach out to us at sales@langchain.dev. Self-hosting LangSmith requires an annual enterprise license that also comes with support and formalized access to the LangChain team.Was this page helpful?NextUser GuideIntroductionInstall LangSmithCreate an API keySetup your environmentLog your first traceCreate your first evaluationNext StepsAdditional ResourcesFAQHow do I migrate projects between organizations?Why aren't my runs aren't showing up in my project?My team deals with sensitive data that cannot be logged. How can I ensure that only my team can access it?CommunityDiscordTwitterGitHubDocs CodeLangSmith SDKPythonJS/TSMoreHomepageBlogLangChain Python DocsLangChain JS/TS DocsCopyright © 2024 LangChain, Inc.", name='langsmith_search', id='f286c7e7-6514-4621-ac60-e4079b37ebe2', tool_call_id='call_FNIgdO97wo51sKx3XZOGLHqT'),
 AIMessage(content="LangSmith is a platform that can significantly aid in testing by offering several features:\n\n1. **Tracing**: LangSmith provides robust tracing capabilities that enable you to monitor your application closely. This feature is particularly useful for tracking the behavior of your application and identifying any potential issues.\n\n2. **Evaluation**: LangSmith allows you to perform comprehensive evaluations of your application. This can help you assess the performance of your application under various conditions and make necessary adjustments to enhance its functionality.\n\n3. **Production Monitoring & Automations**: With LangSmith, you can keep a close eye on your application when it's in active use. The platform provides tools for automatic monitoring and managing routine tasks, helping to ensure your application runs smoothly.\n\n4. **Prompt Hub**: It's a prompt management tool built into LangSmith. This feature can be instrumental when testing various prompts in your application.\n\nOverall, LangSmith helps you build production-grade LLM applications with confidence, providing necessary tools for monitoring, evaluation, and automation.", response_metadata={'token_usage': {'completion_tokens': 200, 'prompt_tokens': 782, 'total_tokens': 982}, 'model_name': 'gpt-4', 'system_fingerprint': None, 'finish_reason': 'stop', 'logprobs': None}, id='run-4b80db7e-9a26-4043-8b6b-922f847f9c80-0')]

Let's take a look at the LangSmith trace to see what is going on under the hood.

Note that the state we get back at the end also contains the tool call and the tool response message.

Now let's try one where it needs to call the search tool:

response = agent_executor.invoke(
    {"messages": [HumanMessage(content="whats the weather in sf?")]}
)
response["messages"]
[HumanMessage(content='whats the weather in sf?', id='e6b716e6-da57-41de-a227-fee281fda588'),
 AIMessage(content='', additional_kwargs={'tool_calls': [{'id': 'call_TGDKm0saxuGKJD5OYOXWRvLe', 'function': {'arguments': '{\n  "query": "current weather in San Francisco"\n}', 'name': 'tavily_search_results_json'}, 'type': 'function'}]}, response_metadata={'token_usage': {'completion_tokens': 23, 'prompt_tokens': 134, 'total_tokens': 157}, 'model_name': 'gpt-4', 'system_fingerprint': None, 'finish_reason': 'tool_calls', 'logprobs': None}, id='run-fd7d5854-2eab-4fca-ad9e-b3de8d587614-0', tool_calls=[{'name': 'tavily_search_results_json', 'args': {'query': 'current weather in San Francisco'}, 'id': 'call_TGDKm0saxuGKJD5OYOXWRvLe'}]),
 ToolMessage(content='[{"url": "https://www.weatherapi.com/", "content": "{\'location\': {\'name\': \'San Francisco\', \'region\': \'California\', \'country\': \'United States of America\', \'lat\': 37.78, \'lon\': -122.42, \'tz_id\': \'America/Los_Angeles\', \'localtime_epoch\': 1714426800, \'localtime\': \'2024-04-29 14:40\'}, \'current\': {\'last_updated_epoch\': 1714426200, \'last_updated\': \'2024-04-29 14:30\', \'temp_c\': 17.8, \'temp_f\': 64.0, \'is_day\': 1, \'condition\': {\'text\': \'Sunny\', \'icon\': \'//cdn.weatherapi.com/weather/64x64/day/113.png\', \'code\': 1000}, \'wind_mph\': 23.0, \'wind_kph\': 37.1, \'wind_degree\': 290, \'wind_dir\': \'WNW\', \'pressure_mb\': 1019.0, \'pressure_in\': 30.09, \'precip_mm\': 0.0, \'precip_in\': 0.0, \'humidity\': 50, \'cloud\': 0, \'feelslike_c\': 17.8, \'feelslike_f\': 64.0, \'vis_km\': 16.0, \'vis_miles\': 9.0, \'uv\': 5.0, \'gust_mph\': 27.5, \'gust_kph\': 44.3}}"}, {"url": "https://www.wunderground.com/hourly/us/ca/san-francisco/94125/date/2024-4-29", "content": "Current Weather for Popular Cities . San Francisco, CA warning 59 \\u00b0 F Mostly Cloudy; Manhattan, NY 56 \\u00b0 F Fair; Schiller Park, IL (60176) warning 58 \\u00b0 F Mostly Cloudy; Boston, MA 52 \\u00b0 F Sunny ..."}]', name='tavily_search_results_json', id='aa0d8c3d-23b5-425a-ad05-3c174fc04892', tool_call_id='call_TGDKm0saxuGKJD5OYOXWRvLe'),
 AIMessage(content='The current weather in San Francisco, California is sunny with a temperature of 64.0°F (17.8°C). The wind is coming from the WNW at a speed of 23.0 mph. The humidity level is at 50%. There is no precipitation and the cloud cover is 0%. The visibility is 16.0 km. The UV index is 5.0. Please note that this information is as of 14:30 on April 29, 2024, according to [Weather API](https://www.weatherapi.com/).', response_metadata={'token_usage': {'completion_tokens': 117, 'prompt_tokens': 620, 'total_tokens': 737}, 'model_name': 'gpt-4', 'system_fingerprint': None, 'finish_reason': 'stop', 'logprobs': None}, id='run-2359b41b-cab6-40c3-b6d9-7bdf7195a601-0')]

We can check out the LangSmith trace to make sure it's calling the search tool effectively.

Streaming Messages​

We've seen how the agent can be called with .invoke to get back a final response. If the agent is executing multiple steps, that may take a while. In order to show intermediate progress, we can stream back messages as they occur.

for chunk in agent_executor.stream(
    {"messages": [HumanMessage(content="whats the weather in sf?")]}
):
    print(chunk)
    print("----")
{'agent': {'messages': [AIMessage(content='', additional_kwargs={'tool_calls': [{'id': 'call_50Kb8zHmFqPYavQwF5TgcOH8', 'function': {'arguments': '{\n  "query": "current weather in San Francisco"\n}', 'name': 'tavily_search_results_json'}, 'type': 'function'}]}, response_metadata={'token_usage': {'completion_tokens': 23, 'prompt_tokens': 134, 'total_tokens': 157}, 'model_name': 'gpt-4', 'system_fingerprint': None, 'finish_reason': 'tool_calls', 'logprobs': None}, id='run-042d5feb-c2cc-4c3f-b8fd-dbc22fd0bc07-0', tool_calls=[{'name': 'tavily_search_results_json', 'args': {'query': 'current weather in San Francisco'}, 'id': 'call_50Kb8zHmFqPYavQwF5TgcOH8'}])]}}
----
{'action': {'messages': [ToolMessage(content='[{"url": "https://www.weatherapi.com/", "content": "{\'location\': {\'name\': \'San Francisco\', \'region\': \'California\', \'country\': \'United States of America\', \'lat\': 37.78, \'lon\': -122.42, \'tz_id\': \'America/Los_Angeles\', \'localtime_epoch\': 1714426906, \'localtime\': \'2024-04-29 14:41\'}, \'current\': {\'last_updated_epoch\': 1714426200, \'last_updated\': \'2024-04-29 14:30\', \'temp_c\': 17.8, \'temp_f\': 64.0, \'is_day\': 1, \'condition\': {\'text\': \'Sunny\', \'icon\': \'//cdn.weatherapi.com/weather/64x64/day/113.png\', \'code\': 1000}, \'wind_mph\': 23.0, \'wind_kph\': 37.1, \'wind_degree\': 290, \'wind_dir\': \'WNW\', \'pressure_mb\': 1019.0, \'pressure_in\': 30.09, \'precip_mm\': 0.0, \'precip_in\': 0.0, \'humidity\': 50, \'cloud\': 0, \'feelslike_c\': 17.8, \'feelslike_f\': 64.0, \'vis_km\': 16.0, \'vis_miles\': 9.0, \'uv\': 5.0, \'gust_mph\': 27.5, \'gust_kph\': 44.3}}"}, {"url": "https://world-weather.info/forecast/usa/san_francisco/april-2024/", "content": "Extended weather forecast in San Francisco. Hourly Week 10 days 14 days 30 days Year. Detailed \\u26a1 San Francisco Weather Forecast for April 2024 - day/night \\ud83c\\udf21\\ufe0f temperatures, precipitations - World-Weather.info."}]', name='tavily_search_results_json', id='d88320ac-3fe1-4f73-870a-3681f15f6982', tool_call_id='call_50Kb8zHmFqPYavQwF5TgcOH8')]}}
----
{'agent': {'messages': [AIMessage(content='The current weather in San Francisco, California is sunny with a temperature of 17.8°C (64.0°F). The wind is coming from the WNW at 23.0 mph. The humidity is at 50%. [source](https://www.weatherapi.com/)', response_metadata={'token_usage': {'completion_tokens': 58, 'prompt_tokens': 602, 'total_tokens': 660}, 'model_name': 'gpt-4', 'system_fingerprint': None, 'finish_reason': 'stop', 'logprobs': None}, id='run-0cd2a507-ded5-4601-afe3-3807400e9989-0')]}}
----

Streaming tokens​

In addition to streaming back messages, it is also useful to be streaming back tokens. We can do this with the .astream_events method.

info

This .astream_events method only works with Python 3.11 or higher.

async for event in agent_executor.astream_events(
    {"messages": [HumanMessage(content="whats the weather in sf?")]}, version="v1"
):
    kind = event["event"]
    if kind == "on_chain_start":
        if (
            event["name"] == "Agent"
        ):  # Was assigned when creating the agent with `.with_config({"run_name": "Agent"})`
            print(
                f"Starting agent: {event['name']} with input: {event['data'].get('input')}"
            )
    elif kind == "on_chain_end":
        if (
            event["name"] == "Agent"
        ):  # Was assigned when creating the agent with `.with_config({"run_name": "Agent"})`
            print()
            print("--")
            print(
                f"Done agent: {event['name']} with output: {event['data'].get('output')['output']}"
            )
    if kind == "on_chat_model_stream":
        content = event["data"]["chunk"].content
        if content:
            # Empty content in the context of OpenAI means
            # that the model is asking for a tool to be invoked.
            # So we only print non-empty content
            print(content, end="|")
    elif kind == "on_tool_start":
        print("--")
        print(
            f"Starting tool: {event['name']} with inputs: {event['data'].get('input')}"
        )
    elif kind == "on_tool_end":
        print(f"Done tool: {event['name']}")
        print(f"Tool output was: {event['data'].get('output')}")
        print("--")
--
Starting tool: tavily_search_results_json with inputs: {'query': 'current weather in San Francisco'}
Done tool: tavily_search_results_json
Tool output was: [{'url': 'https://www.weatherapi.com/', 'content': "{'location': {'name': 'San Francisco', 'region': 'California', 'country': 'United States of America', 'lat': 37.78, 'lon': -122.42, 'tz_id': 'America/Los_Angeles', 'localtime_epoch': 1714427052, 'localtime': '2024-04-29 14:44'}, 'current': {'last_updated_epoch': 1714426200, 'last_updated': '2024-04-29 14:30', 'temp_c': 17.8, 'temp_f': 64.0, 'is_day': 1, 'condition': {'text': 'Sunny', 'icon': '//cdn.weatherapi.com/weather/64x64/day/113.png', 'code': 1000}, 'wind_mph': 23.0, 'wind_kph': 37.1, 'wind_degree': 290, 'wind_dir': 'WNW', 'pressure_mb': 1019.0, 'pressure_in': 30.09, 'precip_mm': 0.0, 'precip_in': 0.0, 'humidity': 50, 'cloud': 0, 'feelslike_c': 17.8, 'feelslike_f': 64.0, 'vis_km': 16.0, 'vis_miles': 9.0, 'uv': 5.0, 'gust_mph': 27.5, 'gust_kph': 44.3}}"}, {'url': 'https://www.weathertab.com/en/c/e/04/united-states/california/san-francisco/', 'content': 'San Francisco Weather Forecast for Apr 2024 - Risk of Rain Graph. Rain Risk Graph: Monthly Overview. Bar heights indicate rain risk percentages. Yellow bars mark low-risk days, while black and grey bars signal higher risks. Grey-yellow bars act as buffers, advising to keep at least one day clear from the riskier grey and black days, guiding ...'}]
--
The| current| weather| in| San| Francisco|,| California|,| USA| is| sunny| with| a| temperature| of| |17|.|8|°C| (|64|.|0|°F|).| The| wind| is| blowing| from| the| W|NW| at| a| speed| of| |37|.|1| k|ph| (|23|.|0| mph|).| The| humidity| level| is| at| |50|%.| [|Source|](|https|://|www|.weather|api|.com|/)|

Adding in memory​

As mentioned earlier, this agent is stateless. This means it does not remember previous interactions. To give it memory we need to pass in a checkpointer. When passing in a checkpointer, we also have to pass in a thread_id when invoking the agent (so it knows which thread/conversation to resume from).

from langgraph.checkpoint.sqlite import SqliteSaver

memory = SqliteSaver.from_conn_string(":memory:")
agent_executor = chat_agent_executor.create_tool_calling_executor(
    model, tools, checkpointer=memory
)

config = {"configurable": {"thread_id": "abc123"}}
for chunk in agent_executor.stream(
    {"messages": [HumanMessage(content="hi im bob!")]}, config
):
    print(chunk)
    print("----")
{'agent': {'messages': [AIMessage(content='Hello Bob! How can I assist you today?', response_metadata={'token_usage': {'completion_tokens': 11, 'prompt_tokens': 131, 'total_tokens': 142}, 'model_name': 'gpt-4', 'system_fingerprint': None, 'finish_reason': 'stop', 'logprobs': None}, id='run-607733e3-4b8d-4137-ae66-8a4b8ccc8d40-0')]}}
----
for chunk in agent_executor.stream(
    {"messages": [HumanMessage(content="whats my name?")]}, config
):
    print(chunk)
    print("----")
{'agent': {'messages': [AIMessage(content='Your name is Bob. How can I assist you further?', response_metadata={'token_usage': {'completion_tokens': 13, 'prompt_tokens': 154, 'total_tokens': 167}, 'model_name': 'gpt-4', 'system_fingerprint': None, 'finish_reason': 'stop', 'logprobs': None}, id='run-e1181ba6-732d-4564-b479-9f1ab6bf01f6-0')]}}
----

Example LangSmith trace

Conclusion​

That's a wrap! In this quick start we covered how to create a simple agent. We've then shown how to stream back a response - not only the intermediate steps, but also tokens! We've also added in memory so you can have a conversation with them. Agents are a complex topic, and there's lot to learn!

For more information on Agents, please check out the LangGraph documentation. This has it's own set of concepts, tutorials, and how-to guides.

Help us out by providing feedback on this documentation page: