Java World Knowledge Sharing Portal: Building Smarter Agents: Integrating Langgraph with REST APIs for Google Gemini

Langgraph, a powerful library for building stateful, multi-actor applications with LLMs, can be even more versatile when integrated with external services. This blog post explores langgraph-restapi, a GitHub repository that provides a seamless way to expose your Langgraph agents as REST APIs, specifically tailored for use with Google Gemini.

What is `langgraph-restapi`?

The langgraph-restapi repository offers a robust REST API wrapper for Langgraph agents, leveraging the power of BentoML and FastAPI. This integration allows your Langgraph-powered agents to interact with other applications and systems through standard RESTful endpoints, supporting both synchronous and streaming interaction modes.

Key Features

RESTful Endpoints: Easily expose your Langgraph agents as web services accessible via HTTP requests.
Synchronous and Streaming Modes: Supports both immediate responses and real-time streaming of agent outputs, providing flexibility for different application needs.
Stateful Agents with Threaded Memory: Maintain conversation context and agent state across multiple interactions, enabling more natural and coherent dialogues.
Flexible Tool Integration: Seamlessly integrate various tools, including Large Language Models (LLMs) like Google Gemini, and custom-built tools to extend the agent’s capabilities.
Deployment Readiness: Built with BentoML and Docker, the solution is designed for easy deployment to various cloud platforms, making it production-ready.

Getting Started

To begin working with langgraph-restapi, you’ll need Python 3.10 or newer, and optionally Docker for containerized deployment. Access to an LLM API (such as Google Gemini) is also required.

Here are the basic steps to get started:

Clone the Repository:
bash git clone https://github.com/shdhumale/langgraph-restapi.git cd langgraph-restapi
Install Requirements:
bash pip install -r requirements.txt
Configure Environment Variables:
Copy the example environment file and update it with your LLM API keys and other configurations.
bash cp .env.example .env
Edit the .env file to provide your specific environment variables.

run this file using this command

python langgraph-restapi.py

This Python script demonstrates how to create an LLM-powered agent using LangGraph and Google’s Gemini model to interact with a REST API (https://api.restful-api.dev/objects) in a conversational way.

Let’s break it down section by section:

1. Import Dependencies

import requests, os, json, re
from langgraph.prebuilt import create_react_agent
from dotenv import load_dotenv
from langchain_google_genai import ChatGoogleGenerativeAI
from typing import List, Dict, Any

requests: For HTTP calls to the REST API.
dotenv: To load the .env file containing the Google API key.
ChatGoogleGenerativeAI: LLM interface to Google’s Gemini via LangChain.
create_react_agent: A prebuilt agent from LangGraph using the ReAct framework (Reasoning + Acting).
typing: For better type hinting.

2. Load Google API Key

load_dotenv()
api_key = os.getenv("GOOGLE_API_KEY")

Loads the API key from a .env file.

3. Initialize LLM (Gemini 2.5 Flash)

llm = ChatGoogleGenerativeAI(
    model="gemini-2.5-flash-preview-05-20",
    google_api_key=api_key
)

Uses Gemini 2.5 Flash (fast inference) to act as the language model for the agent.

4. Define Tool Functions (REST API Wrappers)

Each function below maps to a tool the agent can use:

Function	Purpose
`get_all_objects()`	Fetches all items
`get_objects_by_ids(ids)`	Fetches specific objects
`get_single_object(object_id)`	Fetches one object
`add_object(object_data_json_str)`	Creates a new object
`update_object(object_id, data)`	Full update (PUT)
`patch_object(object_id, data)`	Partial update (PATCH)
`delete_object(object_id)`	Deletes object by ID

5. Create LangGraph Agent

agent = create_react_agent(
    model=llm,
    tools=[...],
    prompt="You are an assistant that manages objects via a REST API."
)

Creates a ReAct-based agent.
The agent can reason through text and call tools (functions) as needed based on user messages.

6. Example Agent Usage

a. Get Multiple Objects

agent.invoke({
    "messages": [{"role": "user", "content": "Get objects for IDs: {1,2,3}"}]
})

b. Get Single Object

agent.invoke({
    "messages": [{"role": "user", "content": "Get single objects for ID: {1}"}]
})

c. Add a New Object

add_object_input = '{"name": "MyTestObject", ...}'
agent.invoke({
    "messages": [{"role": "user", "content": f"Add a new object with data: {add_object_input}"}]
})

Extracts the newly created object’s ID from the agent response.

d. Update the Object (if created)

update_prompt = f"Update the object with ID: ..."
agent.invoke({"messages": [{"role": "user", "content": update_prompt}]})

e. Delete the Object

agent.invoke({
    "messages": [{"role": "user", "content": f"delete the details of object with ID: ..."}]
})

f. Get All Objects

agent.invoke({
    "messages": [{"role": "user", "content": "Get all objects"}]
})

Key Features

You can talk to the REST API via natural language.
LangGraph’s create_react_agent uses LLM + Tool selection to determine the best function to invoke based on the prompt.
Gemini (via langchain_google_genai) is the brain of the system.

Example Prompt to Agent

User: “Add a new object with data: {“name”: “Phone”, “data”: {“year”: 2024}}”
Agent: Calls add_object() and returns the new object ID.

Final Notes

This is a powerful example of combining LLMs with APIs, especially useful in building AI-powered back office agents, API testers, or data managers.
You can replace the Gemini model or API endpoint to suit your needs.

Let me know if you want this turned into a Python notebook, FastAPI app, or LangGraph workflow diagram!

Output:-

(env) C:\vscode-python-workspace\langgraph-restapi>python langgraph-restapi.py
Fetching object with ID: 1
Fetching object with ID: 2
Fetching object with ID: 3
 
--- Result of getting mutiple ID objects: {'messages': [HumanMessage(content='Get objects for IDs: {1,2,3}', additional_kwargs={}, respons
e_metadata={}, id='29277f97-4f44-47db-906a-5f34d327ff7b'), AIMessage(content='', additional_kwargs={'function_call': {'name': 'get_objects
_by_ids', 'arguments': '{"ids": ["1", "2", "3"]}'}}, response_metadata={'prompt_feedback': {'block_reason': 0, 'safety_ratings': []}, 'fin
ish_reason': 'STOP', 'model_name': 'models/gemini-2.5-flash-preview-05-20', 'safety_ratings': []}, id='run--caece8a3-093f-42eb-bfbc-b525c7
9465d2-0', tool_calls=[{'name': 'get_objects_by_ids', 'args': {'ids': ['1', '2', '3']}, 'id': 'c053881a-56f1-4c96-a969-ea5b3109f5f4', 'typ
e': 'tool_call'}], usage_metadata={'input_tokens': 608, 'output_tokens': 25, 'total_tokens': 767, 'input_token_details': {'cache_read': 0}
, 'output_token_details': {'reasoning': 134}}), ToolMessage(content='[{"id": "1", "name": "Google Pixel 6 Pro", "data": {"color": "Cloudy
White", "capacity": "128 GB"}}, {"id": "2", "name": "Apple iPhone 12 Mini, 256GB, Blue", "data": null}, {"id": "3", "name": "Apple iPhone
12 Pro Max", "data": {"color": "Cloudy White", "capacity GB": 512}}]', name='get_objects_by_ids', id='e31ca3ad-25f3-47b4-9500-beb487ba1b53
', tool_call_id='c053881a-56f1-4c96-a969-ea5b3109f5f4'), AIMessage(content='I have retrieved the following objects:\n* Object with ID 1: "
Google Pixel 6 Pro" in "Cloudy White" with "128 GB" capacity.\n* Object with ID 2: "Apple iPhone 12 Mini, 256GB, Blue".\n* Object with ID
3: "Apple iPhone 12 Pro Max" in "Cloudy White" with "512 GB" capacity.', additional_kwargs={}, response_metadata={'prompt_feedback': {'blo
ck_reason': 0, 'safety_ratings': []}, 'finish_reason': 'STOP', 'model_name': 'models/gemini-2.5-flash-preview-05-20', 'safety_ratings': []
}, id='run--5c390692-b7ba-44fb-8cf0-854088e5ee06-0', usage_metadata={'input_tokens': 755, 'output_tokens': 93, 'total_tokens': 942, 'input
_token_details': {'cache_read': 0}, 'output_token_details': {'reasoning': 94}})]}
 
--- Result of getting single objects: {'messages': [HumanMessage(content='Get single objects for ID: {1}', additional_kwargs={}, response_
metadata={}, id='0e2e4532-c15b-441f-be07-e6b7dabd1d6d'), AIMessage(content='', additional_kwargs={'function_call': {'name': 'get_single_ob
ject', 'arguments': '{"object_id": "1"}'}}, response_metadata={'prompt_feedback': {'block_reason': 0, 'safety_ratings': []}, 'finish_reaso
n': 'STOP', 'model_name': 'models/gemini-2.5-flash-preview-05-20', 'safety_ratings': []}, id='run--b6b79445-3e60-4ba7-b489-32d86f11d73d-0'
, tool_calls=[{'name': 'get_single_object', 'args': {'object_id': '1'}, 'id': '7a0b0ba4-da37-4eb2-b890-6a55027e6050', 'type': 'tool_call'}
], usage_metadata={'input_tokens': 605, 'output_tokens': 19, 'total_tokens': 684, 'input_token_details': {'cache_read': 0}, 'output_token_
details': {'reasoning': 60}}), ToolMessage(content='{"id": "1", "name": "Google Pixel 6 Pro", "data": {"color": "Cloudy White", "capacity"
: "128 GB"}}', name='get_single_object', id='1c0956bc-d014-4b32-bae6-1a90c14a59ae', tool_call_id='7a0b0ba4-da37-4eb2-b890-6a55027e6050'),
AIMessage(content='', additional_kwargs={'function_call': {'name': 'get_single_object', 'arguments': '{"object_id": "1"}'}}, response_meta
data={'prompt_feedback': {'block_reason': 0, 'safety_ratings': []}, 'finish_reason': 'STOP', 'model_name': 'models/gemini-2.5-flash-previe
w-05-20', 'safety_ratings': []}, id='run--0945b6ad-54ba-41bd-9486-99353e906bc1-0', tool_calls=[{'name': 'get_single_object', 'args': {'obj
ect_id': '1'}, 'id': '15faacb2-73d5-4b0a-b444-f94912605fc9', 'type': 'tool_call'}], usage_metadata={'input_tokens': 671, 'output_tokens':
19, 'total_tokens': 758, 'input_token_details': {'cache_read': 0}, 'output_token_details': {'reasoning': 68}}), ToolMessage(content='{"id"
: "1", "name": "Google Pixel 6 Pro", "data": {"color": "Cloudy White", "capacity": "128 GB"}}', name='get_single_object', id='78afd91d-8f2
c-491c-ae97-53a49302e9b0', tool_call_id='15faacb2-73d5-4b0a-b444-f94912605fc9'), AIMessage(content='The object with ID 1 is: {"data": {"ca
pacity": "128 GB", "color": "Cloudy White"}, "id": "1", "name": "Google Pixel 6 Pro"}', additional_kwargs={}, response_metadata={'prompt_f
eedback': {'block_reason': 0, 'safety_ratings': []}, 'finish_reason': 'STOP', 'model_name': 'models/gemini-2.5-flash-preview-05-20', 'safe
ty_ratings': []}, id='run--6798fcf3-59b1-4266-a4d9-1d476758799b-0', usage_metadata={'input_tokens': 737, 'output_tokens': 44, 'total_token
s': 862, 'input_token_details': {'cache_read': 0}, 'output_token_details': {'reasoning': 81}})]}
response add_object============= ff8081819782e69e0197a2461eda5668
 
--- Result of adding object (ID): ff8081819782e69e0197a2461eda5668
 
--- Result of updating object: {'messages': [HumanMessage(content='Update the object with ID: \'ff8081819782e69e0197a2461eda5668\' with th
e following data: {"name": "MyTestObject updated", "data": {"year": 2024 , "price": 99.99, "color": "blue updated"}}', additional_kwargs={
}, response_metadata={}, id='8cc3083f-3537-475d-b9b6-42dbeef998fa'), AIMessage(content='', additional_kwargs={'function_call': {'name': 'u
pdate_object', 'arguments': '{"data": {"name": "MyTestObject updated", "data": {"price": 99.99, "year": 2024.0, "color": "blue updated"}},
 "object_id": "ff8081819782e69e0197a2461eda5668"}'}}, response_metadata={'prompt_feedback': {'block_reason': 0, 'safety_ratings': []}, 'fi
nish_reason': 'STOP', 'model_name': 'models/gemini-2.5-flash-preview-05-20', 'safety_ratings': []}, id='run--1e8f4644-36c6-483f-aef5-1b0b7
966dde1-0', tool_calls=[{'name': 'update_object', 'args': {'data': {'name': 'MyTestObject updated', 'data': {'price': 99.99, 'year': 2024.
0, 'color': 'blue updated'}}, 'object_id': 'ff8081819782e69e0197a2461eda5668'}, 'id': '4c6fb529-f43e-4a91-b91b-2bb0fac718e3', 'type': 'too
l_call'}], usage_metadata={'input_tokens': 676, 'output_tokens': 85, 'total_tokens': 851, 'input_token_details': {'cache_read': 0}, 'outpu
t_token_details': {'reasoning': 90}}), ToolMessage(content='{"id": "ff8081819782e69e0197a2461eda5668", "name": "MyTestObject updated", "up
datedAt": "2025-06-24T14:09:57.859+00:00", "data": {"price": 99.99, "year": 2024.0, "color": "blue updated"}}', name='update_object', id='
4aa1c421-f999-4b0e-b338-e0b4c7e28bf7', tool_call_id='4c6fb529-f43e-4a91-b91b-2bb0fac718e3'), AIMessage(content='The object with ID \'ff808
1819782e69e0197a2461eda5668\' has been successfully updated. The updated details are:\n- ID: ff8081819782e69e0197a2461eda5668\n- Name: MyT
estObject updated\n- Data: {"year": 2024, "price": 99.99, "color": "blue updated"}\n- Last updated at: 2025-06-24T14:09:57.859+00:00', add
itional_kwargs={}, response_metadata={'prompt_feedback': {'block_reason': 0, 'safety_ratings': []}, 'finish_reason': 'STOP', 'model_name':
 'models/gemini-2.5-flash-preview-05-20', 'safety_ratings': []}, id='run--ace57c6a-325e-48e7-8826-4cc1d3f55337-0', usage_metadata={'input_
tokens': 877, 'output_tokens': 152, 'total_tokens': 1176, 'input_token_details': {'cache_read': 0}, 'output_token_details': {'reasoning':
147}})]}
 
--- Result of after deleting objects: {'messages': [HumanMessage(content='delete the details of object with ID: ff8081819782e69e0197a2461e
da5668', additional_kwargs={}, response_metadata={}, id='0f5f89af-e34c-4750-b16c-bf7b0bffac2b'), AIMessage(content='', additional_kwargs={
'function_call': {'name': 'delete_object', 'arguments': '{"object_id": "ff8081819782e69e0197a2461eda5668"}'}}, response_metadata={'prompt_
feedback': {'block_reason': 0, 'safety_ratings': []}, 'finish_reason': 'STOP', 'model_name': 'models/gemini-2.5-flash-preview-05-20', 'saf
ety_ratings': []}, id='run--dbf38999-7d36-4e40-a8d5-f42689adf8c3-0', tool_calls=[{'name': 'delete_object', 'args': {'object_id': 'ff808181
9782e69e0197a2461eda5668'}, 'id': '192bef0f-8fba-4727-8176-cc96e1e2f34c', 'type': 'tool_call'}], usage_metadata={'input_tokens': 633, 'out
put_tokens': 45, 'total_tokens': 779, 'input_token_details': {'cache_read': 0}, 'output_token_details': {'reasoning': 101}}), ToolMessage(
content='200', name='delete_object', id='3655c260-0738-4ee3-8034-d18ed47e3602', tool_call_id='192bef0f-8fba-4727-8176-cc96e1e2f34c'), AIMe
ssage(content="The object with ID 'ff8081819782e69e0197a2461eda5668' has been successfully deleted.", additional_kwargs={}, response_metad
ata={'prompt_feedback': {'block_reason': 0, 'safety_ratings': []}, 'finish_reason': 'STOP', 'model_name': 'models/gemini-2.5-flash-preview
-05-20', 'safety_ratings': []}, id='run--ff641837-d12c-4dde-8541-ad0062c6201d-0', usage_metadata={'input_tokens': 695, 'output_tokens': 40
, 'total_tokens': 784, 'input_token_details': {'cache_read': 0}, 'output_token_details': {'reasoning': 49}})]}
 
--- Result of getting all objects: {'messages': [HumanMessage(content='Get all objects', additional_kwargs={}, response_metadata={}, id='4
0ca6531-7560-48bf-846c-8478444839a5'), AIMessage(content='', additional_kwargs={'function_call': {'name': 'get_all_objects', 'arguments':
'{}'}}, response_metadata={'prompt_feedback': {'block_reason': 0, 'safety_ratings': []}, 'finish_reason': 'STOP', 'model_name': 'models/ge
mini-2.5-flash-preview-05-20', 'safety_ratings': []}, id='run--4774bf0f-fce9-44b6-8d93-36027fb56b50-0', tool_calls=[{'name': 'get_all_obje
cts', 'args': {}, 'id': '2705043f-519f-41cd-a940-ecbd3018b9c7', 'type': 'tool_call'}], usage_metadata={'input_tokens': 599, 'output_tokens
': 12, 'total_tokens': 660, 'input_token_details': {'cache_read': 0}, 'output_token_details': {'reasoning': 49}}), ToolMessage(content='[{
"id": "1", "name": "Google Pixel 6 Pro", "data": {"color": "Cloudy White", "capacity": "128 GB"}}, {"id": "2", "name": "Apple iPhone 12 Mi
ni, 256GB, Blue", "data": null}, {"id": "3", "name": "Apple iPhone 12 Pro Max", "data": {"color": "Cloudy White", "capacity GB": 512}}, {"
id": "4", "name": "Apple iPhone 11, 64GB", "data": {"price": 389.99, "color": "Purple"}}, {"id": "5", "name": "Samsung Galaxy Z Fold2", "d
ata": {"price": 689.99, "color": "Brown"}}, {"id": "6", "name": "Apple AirPods", "data": {"generation": "3rd", "price": 120}}, {"id": "7",
 "name": "Apple MacBook Pro 16", "data": {"year": 2019, "price": 1849.99, "CPU model": "Intel Core i9", "Hard disk size": "1 TB"}}, {"id":
 "8", "name": "Apple Watch Series 8", "data": {"Strap Colour": "Elderberry", "Case Size": "41mm"}}, {"id": "9", "name": "Beats Studio3 Wir
eless", "data": {"Color": "Red", "Description": "High-performance wireless noise cancelling headphones"}}, {"id": "10", "name": "Apple iPa
d Mini 5th Gen", "data": {"Capacity": "64 GB", "Screen size": 7.9}}, {"id": "11", "name": "Apple iPad Mini 5th Gen", "data": {"Capacity":
"254 GB", "Screen size": 7.9}}, {"id": "12", "name": "Apple iPad Air", "data": {"Generation": "4th", "Price": "419.99", "Capacity": "64 GB
"}}, {"id": "13", "name": "Apple iPad Air", "data": {"Generation": "4th", "Price": "519.99", "Capacity": "256 GB"}}]', name='get_all_objec
ts', id='2a876e6b-6d9e-4722-ba6b-607a91525ade', tool_call_id='2705043f-519f-41cd-a940-ecbd3018b9c7'), AIMessage(content='Here are all the
objects:\n\n*   **Google Pixel 6 Pro** (ID: 1) - Data: {"capacity": "128 GB", "color": "Cloudy White"}\n*   **Apple iPhone 12 Mini, 256GB,
 Blue** (ID: 2) - No additional data.\n*   **Apple iPhone 12 Pro Max** (ID: 3) - Data: {"capacity GB": 512, "color": "Cloudy White"}\n*
**Apple iPhone 11, 64GB** (ID: 4) - Data: {"color": "Purple", "price": 389.99}\n*   **Samsung Galaxy Z Fold2** (ID: 5) - Data: {"color": "
Brown", "price": 689.99}\n*   **Apple AirPods** (ID: 6) - Data: {"generation": "3rd", "price": 120}\n*   **Apple MacBook Pro 16** (ID: 7)
- Data: {"CPU model": "Intel Core i9", "Hard disk size": "1 TB", "price": 1849.99, "year": 2019}\n*   **Apple Watch Series 8** (ID: 8) - D
ata: {"Case Size": "41mm", "Strap Colour": "Elderberry"}\n*   **Beats Studio3 Wireless** (ID: 9) - Data: {"Color": "Red", "Description": "
High-performance wireless noise cancelling headphones"}\n*   **Apple iPad Mini 5th Gen** (ID: 10) - Data: {"Capacity": "64 GB", "Screen si
ze": 7.9}\n*   **Apple iPad Mini 5th Gen** (ID: 11) - Data: {"Capacity": "254 GB", "Screen size": 7.9}\n*   **Apple iPad Air** (ID: 12) -
Data: {"Capacity": "64 GB", "Generation": "4th", "Price": "419.99"}\n*   **Apple iPad Air** (ID: 13) - Data: {"Capacity": "256 GB", "Gener
ation": "4th", "Price": "519.99"}', additional_kwargs={}, response_metadata={'prompt_feedback': {'block_reason': 0, 'safety_ratings': []},
 'finish_reason': 'STOP', 'model_name': 'models/gemini-2.5-flash-preview-05-20', 'safety_ratings': []}, id='run--7b3af20c-eb87-4f62-9eea-e
0349e865184-0', usage_metadata={'input_tokens': 1141, 'output_tokens': 520, 'total_tokens': 1788, 'input_token_details': {'cache_read': 0}
, 'output_token_details': {'reasoning': 127}})]}
 
(env) C:\vscode-python-workspace\langgraph-restapi>

Lets run the siddhulangraphsingleapi.y

This Python code defines and runs a simple data flow graph using LangGraph, which integrates with LangChain. The purpose of this code is to call a REST API, store its result, and print out part of the response. Here’s a breakdown of what each part does:

Imports

import requests
from langgraph.graph import StateGraph, END
from langchain_core.runnables import RunnableLambda
from typing import TypedDict, Any

requests: Used to make the HTTP API call.
StateGraph, END: LangGraph tools to define a graph and its endpoint.
RunnableLambda: Converts a normal function into a runnable step in LangGraph.
TypedDict: Defines structured types for dictionary-like objects.

Defining Shared State

class GraphState(TypedDict):
    api_response: dict[str, Any]

This class defines the shape of the graph’s shared state.
It has a single key: api_response, which is a dictionary (expected to hold the JSON response from the API).

API Call Function

def call_rest_api(state: GraphState) -> GraphState:
    response = requests.get("https://jsonplaceholder.typicode.com/posts/1")
 
    state = GraphState(api_response={})  # Re-initialize state
    if response.status_code == 200:
        state["api_response"] = response.json()
    else:
        state["api_response"] = {"error": "API call failed"}
 
    return state

Makes a GET request to a fake API (jsonplaceholder.typicode.com).
If successful, stores the response in state["api_response"].
If it fails, stores an error message.

Wrapping in a LangGraph Node

1	`call_api_node = RunnableLambda(call_rest_api)`

Wraps the function so it can be added to a LangGraph as a runnable node.

Building the Graph

builder = StateGraph(GraphState)
builder.add_node("call_api", call_api_node)
builder.set_entry_point("call_api")
builder.add_edge("call_api", END)

StateGraph(GraphState): Initializes a graph with the defined state structure.
add_node: Adds the API call function to the graph.
set_entry_point: Sets "call_api" as the first step.
add_edge: Connects "call_api" node to the END, meaning it’s the only step.

Compiling and Running the Graph

graph = builder.compile()
 
output = graph.invoke(GraphState(api_response={}))
print("Title from call_api:", output.get("api_response", {}).get("title"))
print("API Response:", output)

compile(): Finalizes the graph.
invoke(...): Runs the graph with an empty api_response dictionary.
print(...): Displays the title field and the full API response.

Example Output

If the API call is successful, you might see something like:

Title from call_api: sunt aut facere repellat provident occaecati excepturi optio reprehenderit
API Response: {'api_response': {'userId': 1, 'id': 1, 'title': '...', 'body': '...'}}

Summary

Part	Purpose
`GraphState`	Defines the shape of the graph’s shared data.
`call_rest_api()`	Function to fetch data from a REST API.
`RunnableLambda`	Wraps the function for use in LangGraph.
`StateGraph`	Builds a flowchart-like structure for processing data.
`invoke()`	Runs the graph and collects the final state.

This is a minimal example of how to integrate an external API into a LangGraph workflow.

Output :-python siddhulangraphsingleapi.py

(env) C:\vscode-python-workspace\langgraph-restapi>python siddhulangraphsingleapi.py
Title from call_api: sunt aut facere repellat provident occaecati excepturi optio reprehenderit
API Response: {'api_response': {'userId': 1, 'id': 1, 'title': 'sunt aut facere repellat provident occaecati excepturi optio reprehenderit', 'body': '
quia et suscipit\nsuscipit recusandae consequuntur expedita et cum\nreprehenderit molestiae ut ut quas totam\nnostrum rerum est autem sunt rem eveniet
architecto'}}

Lets run the siddhulangraphmultpileapi.py

This Python code demonstrates how to build a simple two-step workflow using LangGraph that consumes data from two REST APIs in sequence. Let’s go step by step.

Imports

import requests
from langgraph.graph import StateGraph, END
from langchain_core.runnables import RunnableLambda
from typing import TypedDict, Any
from typing import NotRequired

requests: Used to make HTTP GET calls to REST APIs.
langgraph.graph.StateGraph: Defines the flow of computation in a LangGraph.
END: Represents the endpoint of a graph.
RunnableLambda: Wraps a function to be used as a LangGraph step.
TypedDict, Any, NotRequired: Used to define and type the state passed between steps.

Shared Graph State

class GraphState(TypedDict):
    api_response: dict[str, Any]  # Required
    api2_response: NotRequired[dict[str, Any]]  # Optional

This defines the shape of the state passed between graph steps:

api_response: Stores the response from the first API call.
api2_response: Optionally stores response from the second API if a condition is met.

Node 1: Call First API

def call_first_api(state: GraphState) -> GraphState:
    response = requests.get("https://jsonplaceholder.typicode.com/posts/1")
    new_state = state.copy()
    if response.status_code == 200:
        new_state["api_response"] = response.json()
    else:
        new_state["api_response"] = {"error": "API call failed"}
    return new_state

Calls a mock API and fetches post #1.
Adds the result to api_response in the state.

Node 2: Call Second API (Conditional)

def call_second_api(state: GraphState) -> GraphState:
    title = state.get('api_response', {}).get('title')
    print(f"The fetched title is: {title}")
    new_state = state.copy()
 
    if title and "sunt aut facere repellat provident occaecati excepturi optio reprehenderit".lower() in title.lower():
        response = requests.get("https://jsonplaceholder.typicode.com/posts/2")
        if response.status_code == 200:
            new_state["api2_response"] = response.json()
        else:
            new_state["api2_response"] = {"error": "API 2 failed"}
    return new_state

Extracts the title from the first API response.
If the title contains a specific phrase, it calls a second API (post #2).
Adds result to api2_response in the state.

Build the LangGraph

step_1 = RunnableLambda(call_first_api)
step_2 = RunnableLambda(call_second_api)
 
builder = StateGraph(GraphState)
builder.add_node("call_api_1", step_1)
builder.add_node("call_api_2", step_2)
 
builder.set_entry_point("call_api_1")
builder.add_edge("call_api_1", "call_api_2")
builder.add_edge("call_api_2", END)
 
graph = builder.compile()

This defines a graph with two steps:

Start at call_api_1
Then run call_api_2
Then end

Run the Graph

output = graph.invoke(GraphState(api_response={}))
print("Title from API 1:", output.get("api_response", {}).get("title"))
print("Response from API 2:", output.get("api2_response"))

Executes the LangGraph from an empty initial state.
Prints the title fetched from API 1.
Prints the response from API 2 (if fetched).

Summary

Step	What It Does
1	Calls `https://jsonplaceholder.typicode.com/posts/1` and stores the response in `api_response`.
2	If the `title` from API 1 matches a specific string, it calls API 2 (`/posts/2`) and stores it in `api2_response`.
3	LangGraph manages the flow between steps using nodes and edges.

Let me know if you’d like this converted into a reusable module or integrated with your own APIs.

Output :-python siddhulangraphmultpileapi.py

(env) C:\vscode-python-workspace\langgraph-restapi>python siddhulangraphmultpileapi.py
The fetched title is: sunt aut facere repellat provident occaecati excepturi optio reprehenderit
Title from API 1: sunt aut facere repellat provident occaecati excepturi optio reprehenderit
Response from API 2: {'userId': 1, 'id': 2, 'title': 'qui est esse', 'body': 'est rerum tempore vitae\nsequi sint nihil reprehenderit dolor beatae ea dolores ne
que\nfugiat blanditiis voluptate porro vel nihil molestiae ut reiciendis\nqui aperiam non debitis possimus qui neque nisi nulla'}

This project is licensed under the MIT License, making it open for use and contribution.

By using langgraph-restapi, developers can unlock new possibilities for integrating intelligent agents into their existing systems, building more dynamic and responsive applications with Google Gemini and Langgraph.

For more details and to explore the code, visit the langgraph-restapi GitHub repository.

Java World Knowledge Sharing Portal

Thursday, June 26, 2025

Building Smarter Agents: Integrating Langgraph with REST APIs for Google Gemini

What is `langgraph-restapi`?

Key Features

Getting Started

1. Import Dependencies

2. Load Google API Key

3. Initialize LLM (Gemini 2.5 Flash)

4. Define Tool Functions (REST API Wrappers)

5. Create LangGraph Agent

6. Example Agent Usage

a. Get Multiple Objects

b. Get Single Object

c. Add a New Object

d. Update the Object (if created)

e. Delete the Object

f. Get All Objects

Key Features

Example Prompt to Agent

Final Notes

Imports

Defining Shared State

API Call Function

Wrapping in a LangGraph Node

Building the Graph

Compiling and Running the Graph

Example Output

Summary

Imports

Shared Graph State

Node 1: Call First API

Node 2: Call Second API (Conditional)

Build the LangGraph

Run the Graph

Summary

No comments:

Total Pageviews

Blog Archive

Contributors

Java World Knowledge Sharing Portal

Thursday, June 26, 2025

Building Smarter Agents: Integrating Langgraph with REST APIs for Google Gemini

What is langgraph-restapi?

Key Features

Getting Started

1. Import Dependencies

2. Load Google API Key

3. Initialize LLM (Gemini 2.5 Flash)

4. Define Tool Functions (REST API Wrappers)

5. Create LangGraph Agent

6. Example Agent Usage

a. Get Multiple Objects

b. Get Single Object

c. Add a New Object

d. Update the Object (if created)

e. Delete the Object

f. Get All Objects

Key Features

Example Prompt to Agent

Final Notes

Imports

Defining Shared State

API Call Function

Wrapping in a LangGraph Node

Building the Graph

Compiling and Running the Graph

Example Output

Summary

Imports

Shared Graph State

Node 1: Call First API

Node 2: Call Second API (Conditional)

Build the LangGraph

Run the Graph

Summary

No comments:

Total Pageviews

Blog Archive

Contributors

What is `langgraph-restapi`?