aops 0.5.0

Python SDK for AOps — prompt version management

aops

Python SDK for AOps — prompt version management and agent tracing platform.

Provides a framework-agnostic pull() that works with any LLM SDK, a run() context manager for execution tracing, and a LangChain integration.

Installation

pip install aops

With LangChain integration:

pip install "aops[langchain]"

Quick Start

Pull a prompt

import aops
from aops import pull
from openai import OpenAI

aops.init(api_key="aops_...", agent="my-agent")

system_prompt = pull("my-chain")  # agent resolved from init()

client = OpenAI()
response = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[
        {"role": "system", "content": system_prompt},
        {"role": "user", "content": "Hello!"},
    ],
)
print(response.choices[0].message.content)

Record execution traces

Wrap your agent logic in aops.run() to automatically record which chains were called, in what order, and with what latency. The data is posted to the AOps backend and visualized in the Flow tab.

import aops

aops.init(api_key="aops_...", agent="my-agent")

# Chains pulled outside run() are not traced (good for shared system prompts)
system_prompt = aops.pull("system")

def handle(inquiry: str) -> str:
    with aops.run():
        classify_prompt = aops.pull("classify")      # traced
        category = classify(classify_prompt, inquiry)

        response_prompt = aops.pull(f"respond-{category}")  # traced
        return respond(system_prompt, response_prompt, inquiry)

On block exit, the SDK posts started_at, ended_at, and the ordered list of chain calls to POST /agents/:id/runs. If the backend is unreachable, a warning is logged and the exception is suppressed — your agent is never interrupted by a tracing failure.

Anthropic SDK

import aops
from aops import pull
from anthropic import Anthropic

aops.init(api_key="aops_...", agent="my-agent")

system_prompt = pull("my-chain")

client = Anthropic()
message = client.messages.create(
    model="claude-haiku-4-5-20251001",
    max_tokens=1024,
    system=system_prompt,
    messages=[{"role": "user", "content": "Hello!"}],
)
print(message.content[0].text)

LangChain

import aops
from aops.langchain import pull
from langchain_core.prompts import ChatPromptTemplate, HumanMessagePromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain_openai import ChatOpenAI

aops.init(api_key="aops_...", agent="my-agent")

prompt = pull("my-chain")

chain = (
    ChatPromptTemplate.from_messages([
        prompt,
        HumanMessagePromptTemplate.from_template("{user_input}"),
    ])
    | ChatOpenAI(model="gpt-4o-mini")
    | StrOutputParser()
)

result = chain.invoke({"user_input": "Hello!"})

Requirements

• Python 3.12+
• AOps backend running (self-hosted)
• API key issued from the AOps UI: Agent detail page → API Keys → New API Key

Examples

examples/
  openai_example.py     raw pull() + OpenAI SDK
  anthropic_example.py  raw pull() + Anthropic SDK
  langchain_example.py  aops.langchain — pull(), @chain_prompt
  live_updates.py       background polling / live update detection

Docs

Guide	Description
Configuration	API key, aops.init(), environment variables
API Reference	pull(), aops.langchain.pull(), @chain_prompt
Live Updates	Polling, pattern selection guide
Run Tracing	aops.run(), how traces are recorded and posted, async safety
LangChain Compatibility	Class-based vs LCEL, RunnableLambda lazy-pull pattern

License

MIT