Python SDK

pip install cloudmind

Monitoring wrappers

from openai import OpenAI
from cloudmind.openai import monitor_openai

client = monitor_openai(OpenAI())

Decorator

from cloudmind import trace

@trace(name="answer-question")
def answer(question: str) -> str:
    chunks = retrieve(question)
    return generate(question, chunks)

The decorator works on both sync and async functions.

Context manager

from cloudmind import cloudmind

with cloudmind.trace(name="answer-question", user_id="u_123") as t:
    with t.span(name="retrieve", kind="retrieval") as span:
        docs = store.search(question, k=5)
        span.record(query=question, chunks=docs)

    answer = generate(question, docs)

Prompts

prompt = cloudmind.prompts.get("support-summary", label="production")
messages = prompt.render(ticket=ticket.body, tone="concise")

Async

Every method has an async equivalent prefixed with a:

await cloudmind.atrace(...)
await cloudmind.aflush()

Flushing

cloudmind.flush()

Call before exiting a Lambda handler or a short-lived script.