Skip to content

LLM interaction

Script in langroid-examples

A full working example for the material in this section is in the script in the langroid-examples repo: examples/quick-start/

Let's start with the basics -- how to directly interact with an OpenAI LLM using Langroid.

Configure, instantiate the LLM class

First define the configuration for the LLM, in this case one of the OpenAI GPT chat models:

import langroid as lr

cfg = lr.language_models.OpenAIGPTConfig(

About Configs

A recurring pattern you will see in Langroid is that for many classes, we have a corresponding Config class (an instance of a Pydantic BaseModel), and the class constructor takes this Config class as its only argument. This lets us avoid having long argument lists in constructors, and brings flexibility since adding a new argument to the constructor is as simple as adding a new field to the corresponding Config class. For example the constructor for the OpenAIGPT class takes a single argument, an instance of the OpenAIGPTConfig class.

Now that we've defined the configuration of the LLM, we can instantiate it:

mdl = lr.language_models.OpenAIGPT(cfg)

We will use OpenAI's GPT4 model's chat completion API.

Messages: The LLMMessage class

This API takes a list of "messages" as input -- this is typically the conversation history so far, consisting of an initial system message, followed by a sequence of alternating messages from the LLM ("Assistant") and the user. Langroid provides an abstraction LLMMessage to construct messages, e.g.

from langroid.language_models import Role, LLMMessage

msg = LLMMessage(
    content="what is the capital of Bangladesh?", 

LLM response to a sequence of messages

To get a response from the LLM, we call the mdl's chat method, and pass in a list of messages, along with a bound on how long (in tokens) we want the response to be:

messages = [
    LLMMessage(content="You are a helpful assistant", role=Role.SYSTEM), #(1)!
    LLMMessage(content="What is the capital of Ontario?", role=Role.USER), #(2)!

response =, max_tokens=200)

  1. 🙋‍♂️ With a system message, you can assign a "role" to the LLM
  2. 🙋‍♂️ Responses from the LLM will have role Role.ASSISTANT; this is done behind the scenes by the response.to_LLMMessage() call below.

The response is an object of class LLMResponse, which we can convert to an LLMMessage to append to the conversation history:


You can put the above in a simple loop, to get a simple command-line chat interface!

from rich import print
from rich.prompt import Prompt #(1)!

messages = [
    LLMMessage(role=Role.SYSTEM, content="You are a helpful assitant"),

while True:
    message = Prompt.ask("[blue]Human")
    if message in ["x", "q"]:
    messages.append(LLMMessage(role=Role.USER, content=message))

    response =, max_tokens=200)
    print("[green]Bot: " + response.message)
  1. Rich is a Python library for rich text and beautiful formatting in the terminal. We use it here to get a nice prompt for the user's input. You can install it with pip install rich.

See examples/quick-start/ for a complete example that you can run using

python3 examples/quick-start/

Here is a screenshot of what it looks like:


Next steps

You might be thinking: "It is tedious to keep track of the LLM conversation history and set up a loop. Does Langroid provide any abstractions to make this easier?"

We're glad you asked! And this leads to the notion of an Agent. The next section will show you how to use the ChatAgent class to set up a simple chat Agent in a couple of lines of code.