First Example

Import Council.

from council.chains import Chain
from council.skills import LLMSkill
from council.llm import OpenAILLM

Setup API keys in .env file (example in repository) and use it to setup the LLM (here: OpenAILLM).

import dotenv
import os

dotenv.load_dotenv()
print(os.getenv("OPENAI_API_KEY", None) is not None)

openai_llm = OpenAILLM.from_env()

Create your first Hello World Skill and Wrap it in a Chain.

prompt = "You are responding to every prompt with a short poem titled hello world"
hw_skill = LLMSkill(llm=openai_llm, system_prompt=prompt)
hw_chain = Chain(name="Hello World", description="Answers with a poem titled Hello World", runners=[hw_skill])

Create a second Skill (that responds only with Emojis).

prompt = "You are responding to every prompt with an emoji that best addresses the question asked or statement made"
em_skill = LLMSkill(llm=openai_llm, system_prompt=prompt)
em_chain = Chain(name="Emoji Agent", description="Responds to every prompt with an emoji that best fits the prompt",
                 runners=[em_skill])

Create a Controller to route prompts to chains. Here we use the straight-forward LLMController in which an LLM instance is tasked to make a routing decision.

from council.controllers import LLMController

controller = LLMController(chains=[hw_chain, em_chain], llm=openai_llm, response_threshold=5)

Create an Evaluator. Here, we use an LLMEvaluator in which an LLM is tasked to evaluate each response received.

from council.evaluators import LLMEvaluator

evaluator = LLMEvaluator(llm=openai_llm)

Finalize setup of the Hello World first Agent by combining all components created.

from council.filters import BasicFilter
from council.agents import Agent

agent = Agent(controller=controller, evaluator=evaluator, filter=BasicFilter())

Now, we are ready to invoke the agent.

result = agent.execute_from_user_message("hello world?!")
print(result.best_message.message)

result = agent.execute_from_user_message("Represent with emojis, council a multi-agent framework")
print(result.best_message.message)