Skip to main content

Get Started with Python

This guide will walk you through creating and deploying your first AI agent using the aiXplain Python SDK. You'll learn how to specify a Large Language Model (LLM), equip the agent with tools, and integrate the agent into your application.

Setup​

Set an aiXplain API Key​

Generate an API key from the API keys page. Once generated, you can use it in one of two ways:

  1. Export it as an environment variable in your terminal or
  2. Set it in your Python project using either environment variables or a .env file with python-dotenv.
export AIXPLAIN_API_KEY="<YOUR_API_KEY>"
import os
os.environ["AIXPLAIN_API_KEY"] = "<YOUR_API_KEY>"

Install the aiXplain SDK​

pip install aixplain

Create an Agent​

Agents are defined using a name, description, instructions and any tools you equip the agent with. The agent below is eqipped with Open Weather API tool. Find more in aiXplain's marketplace.

from aixplain.factories import AgentFactory
from aixplain.modules.agent.tool.model_tool import ModelTool

Tool = "66f83c216eb563266175e201"  #Open Weather API tool

weather_agent = AgentFactory.create(
    name="Weather Agent",
    description="Agent for weather updates.",
    instructions="An agent that answers queries about the current weather.",
    tools=[
		ModelTool(model=Tool),
	],

)

Only the instructions field influences an agent's behavior. The name and description are metadata and do not affect execution.

note

aiXplain agents are currently optimized for OpenAI (GPT-4o Mini default) and Meta models. However, you can override the default llm with your preferred model from a collection of 170+ LLMs in the aiXplain marketplace.

Run the agent​

response = weather_agent.run("What is the weather in Liverpool, UK?")
response.data.output
Show output

Enable conversational memory​

Retrieve the session_id from the first query and include it in subsequent requests to maintain the agent’s conversation history.

session_id = response.data.session_id
response = weather_agent.run("Is that too cold for shorts?", session_id=session_id)
response.data.output
Show output

Deploy the agent​

Creating an agent generates a draft agent with a temporary endpoint URL that expires after 24 hours. Draft agents are ideal for development and debugging.

To make an agent permanent, you need to deploy it. By default, agents are deployed on aiXplain’s multi-tenant AWS cloud environment.

weather_agent.deploy()

Create a Team Agent​

A Team Agent orchestrates multiple agents to execute multi-step plans collaboratively.

Add tools​

The agentic framework supports different types of tools. This section demonstrates how to list, try and add an AI model as an agent tool.

List models​

from aixplain.factories import ModelFactory
from aixplain.enums import Function

# List text generation models 
model_list = ModelFactory.list(function=Function.TEXT_GENERATION, page_size=50)["results"]

for model in model_list:
  print(model.id, model.name, model.supplier)
Show output

Alternatively, browse models and other assets in the marketplace.

Try the model​

In this example, we will use the Google Search by Scale SERP model.

from aixplain.factories import ModelFactory

scrape_model = ModelFactory.get("66f423426eb563fa213a3531") # Scraper Website Tool
result = scrape_model.run("https://www.bbc.com/news")
display(result.data)
Show output

Create a Scraper Agent​

Create a scraper agent to scrape websites.

# Web scraper agent
from aixplain.factories import AgentFactory

scraper_agent = AgentFactory.create(
    name="Scraper Agent",
    description="An agent that answers queries using website scraping.",
    instructions="Scrapes websites to extract information",
    tools=[AgentFactory.create_model_tool(
        model=scrape_model,
        description="The input query of this tool must be of the form 'text': 'https://website.com'.")]
)

print(scraper_agent.name, scraper_agent.id)
Show output

Run and Deploy the Agent​

from aixplain.modules.agent import OutputFormat

response = scraper_agent.run("What are the latest interesting news listed at https://www.cnn.com/'?", output_format=OutputFormat.MARKDOWN)
print(response.data["output"])
Show output

Deploy the agent

scraper_agent.deploy()

Create a Web Wiki Agent​

Create a web wiki agent to answer questions using wikipedia

from aixplain.modules.agent.tool.model_tool import ModelTool

# Wiki agent
wiki_agent = AgentFactory.create(
    name="Wiki Agent",
    description="An agent that answers queries using wikipedia.",
    instructions="Queries wikipedia to answer user questions",
    tools=[ModelTool(model="6633fd59821ee31dd914e232")] # Wikipedia utility
)

print(wiki_agent.name, wiki_agent.id)
Show output

Run and Deploy the Agent​

response = wiki_agent.run("How is the friendship paradox described?")
print(response.data["output"])
Show output

Deploy the agent

wiki_agent.deploy()

Create a Team Agent​

Create a Web and Wiki Team Agent with the scraper agent and wiki agent.

from aixplain.factories import TeamAgentFactory

team = TeamAgentFactory.create(
    name="Wiki and Web Team Agent",
    description="You take user queries and search them using wiki or by web scraping URLs if appropriate.",
    instructions="You take user queries and search them using wiki or by web scraping URLs if appropriate.",
    agents=[scraper_agent, wiki_agent],
    llm_id="6646261c6eb563165658bbb1" # GPT-4o
)


print(team.__dict__) # Team agent attributes

# Run the team agent for the first time
result = team.run("Tell me about OpenAI. They have a website, https://openai.com/.") # run team agent
print(result['data']['output'])

# Use the same `session ID` to maintain conversational context
session = result["data"]["session_id"]
result = team.run("Who is the founder?",session_id=session)
print(result['data']['output'])

# Deploy the team agent to make it production-ready
team.deploy()
Show output

Next – Customize your agent​

Now that you’ve seen how agents can collaborate as a team, let’s dive deeper into how you can customize individual agents even further. From refining instructions and selecting tools to configuring memory, guardrails, and model behavior—agents on aiXplain are highly adaptable to your specific use case.

Learn more about agents →