opperai

Opper Python SDK

This is the Opper Python SDK. See below for getting started, and the docs for more information. The SDK has builtin documentation and examples in function docstrings, which should be visible in your code editor as you are using the functions.

Install

pip install opperai

Configuration

Environment variable

• OPPER_API_KEY environment variable is read by the SDK if no api_key is provided to the Client object.

Using opper

from typing import List

from opperai import Index, Opper, trace
from opperai.types import DocumentIn
from opperai.types.indexes import DocumentIn, RetrievalResponse
from pydantic import BaseModel


class Answer(BaseModel):
    steps: List[str]


class QuestionAndContext(BaseModel):
    question: str
    context: List[RetrievalResponse]


@trace
def answer_question(index: Index, question: str) -> Answer:
    results = index.query(question, 1, None)

    result, response = opper.call(
        name="answer_question",
        instructions="Answer the question and provide the steps to do so",
        input=QuestionAndContext(question=question, context=results),
        output_type=Answer,
    )
    response.span.save_metric("artificial_score", 5)

    return result


@trace
def translate(answer: Answer, language: str) -> str:
    result, _ = opper.call(
        name="translate",
        instructions="Translate the answer to the given language",
        input=answer,
        output_type=Answer,
    )
    return result


qna = [
    {
        "question": "I cannot log in to my account",
        "answer": "Use the reset password feature by clicking on 'Forgot password?' and then follow the instructions from email",
        "id": "1",
    },
    {
        "question": "How can I see my invoices?",
        "answer": "Go to the billing section and click on 'Invoices'",
        "id": "2",
    },
    {
        "question": "How can I add a new user to my account?",
        "answer": "Upgrade account and add the user to your account",
        "id": "3",
    },
]

opper = Opper()


def index_qna(qnas: list[dict]):
    index = opper.indexes.create("qna")

    for qna in qnas:
        index.add(
            DocumentIn(
                key=qna["id"],
                content=f"question: {qna['question']}\nanswer: {qna['answer']}",
                metadata={
                    "id": qna["id"],
                },
            )
        )

    return index


def run():
    index = index_qna(qna)

    question = "How can I see my invoices?"

    with opper.traces.start("answer_question") as trace:
        answer = answer_question(index, question)
        print(answer)


run()

More examples

See examples in our documentation and examples folder.