Getting Started
There are two ways to use Agora: as a sender agent (i.e. a client) or as a receiver agent (i.e. a server). An agent can also act as both a sender and a receiver.
In this quick tutorial, we'll establish a structured communication channel between two agents:
- A LangChain agent that wants to retrieve temperature data
- A Camel agent that has access to weather data
Installation
pip install agora-protocol
For this tutorial, you'll also want to install the two agent frameworks:
pip install langchain_openai
pip install Pillow requests_oauthlib # Dependencies for camel-ai
pip install camel-ai
We'll use an OpenAI model as base. You can set the API key via the OPENAI_API_KEY= environmental variable.
Sender
The Sender is an agent designed to execute tasks. Defining a task is as simple as taking a documented Python function and adding the @sender.task decorator.
import agora
from langchain_openai import ChatOpenAI
model = ChatOpenAI(model="gpt-4o-mini")
toolformer = agora.toolformers.LangChainToolformer(model)
sender = agora.Sender.make_default(toolformer)
@sender.task()
def get_temperature(city : str) -> int:
"""
Get the temperature for a given city.
Parameters:
city: The name of the city for which to retrieve the weather
Returns:
The temperature in °C for the given city.
"""
pass
Note: any properly annotated function with Google-style docstrings can be automatically converted to a task. Refer to this page for other ways to describe tasks.
The function is automatically converted to a task function. A task function takes exactly the same arguments, in addition to a keyword-only argument target which represents the address of the remote agent.
response = get_temperature('New York', target='http://localhost:5000')
print(response) # Output: 25
When running this code, the Sender agent will begin a conversation with the Receiver agent on localhost:5000. The two will exchange information first using natural language and then, depending on the need, with structured data and automatic routines (see [specification]). All of this is abstracted away and happens under the hood.
Receiver
Let's now setup a Receiver instance on port 5000. This time, we'll use a Camel agent with one tool, weather_db:
import agora
import camel.types # Needs to be installed separately
toolformer = agora.toolformers.CamelToolformer(
camel.types.ModelPlatformType.OPENAI,
camel.types.ModelType.GPT_4O
)
def weather_db(city: str) -> dict:
"""Gets the temperature and precipitation in a city.
Args:
city: The name of the city for which to retrieve the weather
Returns:
A dictionary containing the temperature and precipitation in the city (both ints)
"""
# Put your tool logic here
return {
'temperature': 25,
'precipitation': 12
}
receiver = agora.Receiver.make_default(toolformer, tools=[weather_db])
A receiver needs to be wrapped in a server capable of handling HTTP queries. For convinience's sake, the Agora client provides a Flask server that does that out of the box:
server = agora.ReceiverServer(receiver)
server.run(port=5000)
We're done! The get_temperature task can now be used seamlessly in your custom workflow, or even by another agent.