Skip to main content
Nightly lets your agents learn and improve over time by autonomously and asynchronously deciding what interactions are worth remembering, and then efficiently accessing those memories at inference time. The memory_group is a unique key for a persistent pool of memories (e.g., one per user or per agent type). The user_goal is a hint you can provide to help your agent better curate which interactions are worth saving as memories. 
from openai import OpenAI

client = OpenAI(
    base_url="https://rkdune--symmetry.modal.run/v1/",
    api_key="YOUR_ASYMMETRIC_API_KEY",  # get an API key from the dashboard
)

completion = client.chat.completions.create(
    model="openai/gpt-4o-mini",  # use any model
    messages=[{"role": "user", "content": "Explain the Sylow theorems."}],
    stream=False,
    extra_body={
        "nightly": ["my_math_agent", "This agent helps prove theorems in abstract algebra."]
        #            ^ memory_group      ^ user_goal
    }
)