Nebuly SDK

The SDK for instrumenting applications for tracking AI costs.

Setup

To set up the code quality checks for this project:

1. Clone the repository
2. Run the setup command to install the necessary requirements, including Poetry for handling dependencies

make setup

Code Formatting and Linting

The code formatting and linting checks help maintain consistent style and identify potential issues. Black and Ruff are automatically invoked with each commit, but they can also be utilized independently without committing changes:

• To display the issues detected by the linter

make lint

• To automatically apply the formatter changes and the suggested changes by the linter, use the following command

make lint-fix

Supported Providers

- OpenAI
- Azure OpenAI
- Cohere
- Anthropic
- HuggingFace pipelines
- HuggingFace HUB
- LangChain
- LlamaIndex
- Amazon Bedrock
- Amazon SageMaker
- Google PALM API
- Google VertexAI

Usage

Make sure you initialize Nebuly before importing other libraries like openai, cohere, huggingface, etc.

Simple usage

In the simple case, you can just import nebuly and call the init function with your API key. This will automatically setup all the tracking for you. After that, you can call the other libraries as normal.

Example with OpenAI

import os
import nebuly

api_key = os.getenv("NEBULY_API_KEY")
nebuly.init(api_key=api_key)

import os
from openai import OpenAI

client = OpenAI()
chat_completion = client.chat.completions.create(
    messages=[
        {
            "role": "user",
            "content": "Say this is a test",
        }
    ],
    model="gpt-3.5-turbo",
    user_id="user-123",
    feature_flags=["new-feature_flag"],
)

Advanced usage: Context managers

In the simple case, each call will be stored as a separate Interaction, you can use context managers to group more calls in a single Interaction:

Example with OpenAI and Cohere

import os
import nebuly
from nebuly.contextmanager import new_interaction

api_key = os.getenv("NEBULY_API_KEY")
nebuly.init(api_key=api_key)

# Setup OpenAI
import openai

openai.api_key = os.getenv("OPENAI_API_KEY")

# Setup Cohere
import cohere

co = cohere.Client(os.getenv("COHERE_API_KEY"))

with new_interaction(user_id="test_user", user_group_profile="test_group") as interaction:
    # interaction.set_input("Some custom input")
    # interaction.set_history([{"role": "user/assistant", "content": "sample content"}}])
    completion_1 = openai.ChatCompletion.create(
        model="gpt-3.5-turbo",
        messages=[
            {"role": "system", "content": "You are an helpful assistant"},
            {"role": "user", "content": "Hello world"}
        ]
    )
    completion_2 = co.generate(
        prompt='Please explain to me how LLMs work',
    )
    # interaction.set_output("Some custom output")

LangChain Callbacks

import os

from langchain.chains import LLMChain
from langchain.chat_models import ChatOpenAI
from langchain.prompts import PromptTemplate
from nebuly.providers.langchain import LangChainTrackingHandler

callback = LangChainTrackingHandler(
    user_id="test_user",
    api_key=os.getenv("NEBULY_API_KEY"),
)

llm = ChatOpenAI(temperature=0.9)
prompt = PromptTemplate(
    input_variables=["product"],
    template="What is a good name for a company that makes {product}?",
)

chain = LLMChain(llm=llm, prompt=prompt)
result = chain.run(
    "colorful socks",
    callbacks=[callback],
)

LlamaIndex Callbacks

import os
from nebuly.providers.llama_index import LlamaIndexTrackingHandler

handler = LlamaIndexTrackingHandler(
    api_key=os.getenv("NEBULY_API_KEY"), user_id="test_user"
)

import llama_index
from llama_index import SimpleDirectoryReader, VectorStoreIndex

llama_index.global_handler = handler

documents = SimpleDirectoryReader("data").load_data()
index = VectorStoreIndex.from_documents(documents)
query_engine = index.as_query_engine()
response = query_engine.query("What did the author do growing up?")

Variants Usage

from nebuly.ab_testing import ABTesting

client = ABTesting("your_nebuly_api_key")

variants = client.get_variants(
  user="<user_id>",
  feature_flags=["feature_flag_a", "feature_flag_b"]
)
print(variants)