Quick start

It takes less than 2 minutes to track interactions to Nebuly. In this guide, we provide some guidance on what to track and how to track it.

We recommend starting with just one interaction to gain an immediate understanding of the product.

Step 1: Create a project and API Key

Step 2: Track your first interaction

Use your LLMs as usual, just add:

Nebuly’s API key;
user_id within the model call.

Python Example

import nebuly
import openai

nebuly.init(api_key="<nebuly_api_key>")
openai.api_key = "<your_openai_api_key>"

response = openai.ChatCompletion.create(
    model="gpt-3.5-turbo",
    messages=[
        {
            "role": "system",
            "content": "You are an helpful assistant"
        },
        {
            "role": "user",
            "content": "Hello, I need help with my computer"
        }
    ],
    user_id="test_user",
)

Python Example

import nebuly
import openai

nebuly.init(api_key="<nebuly_api_key>")
openai.api_key = "<your_openai_api_key>"

response = openai.ChatCompletion.create(
    model="gpt-3.5-turbo",
    messages=[
        {
            "role": "system",
            "content": "You are an helpful assistant"
        },
        {
            "role": "user",
            "content": "Hello, I need help with my computer"
        }
    ],
    user_id="test_user",
)

Python Example

import nebuly

nebuly.init(api_key="{nebuly_api_key}")

from anthropic import Anthropic, HUMAN_PROMPT, AI_PROMPT

anthropic = Anthropic(api_key="<anthropic_api_key>")
completion = anthropic.completions.create(
    model="claude-2",
    max_tokens_to_sample=300,
    prompt=f"{HUMAN_PROMPT} How many toes do dogs have?{AI_PROMPT}",
    user_id="test_user"
)

Python Example

import nebuly
import cohere

nebuly.init(api_key="{nebuly_api_key}")
client = cohere.Client("<cohere_api_key>")

response = client.chat(
    message="Hello, I need help with my computer",
    chat_history=[],
    model="command",
    user_id="test_user"
)

Python Example

import nebuly
nebuly.init(api_key="{nebuly_api_key}")

from huggingface_hub import InferenceClient

# Define the model name
model_name = "mistralai/Mistral-7B-Instruct-v0.1"

# Define the input data
input_data = "Hello, how are you doing today?"

# Load the model
client = InferenceClient(model_name, token="<your_api_key>")
output_data = client.text_generation(
    input_data,
    user_id="test_user"
)

Python Example

import nebuly
from langchain.chains import LLMChain
from langchain.llms import OpenAI
from langchain.prompts import PromptTemplate

nebuly.init(api_key="{nebuly_api_key}")


prompt = PromptTemplate.from_template("You are an helpful assistant. You must answer to the following question.\n{question}")
llm = OpenAI(
    openai_api_key="<openai_api_key>", 
    model="gpt-3.5-turbo-instruct"
)
chain = LLMChain(llm=llm, prompt=prompt)

response = chain.run(
    question="What is the capital of France?",
    user_id="test_user"
)

For use cases not using a third party SDK or for situations where an highly customization is needed we also expose the raw endpoint you can use to send the interaction. Below an example of integrating an openai call with the custom endpoint.

Python Example

import request
from datetime import datetime, timezone
import openai


NEBULY_API_KEY = "your_nebuly_api_key"
openai.api_key = "<your_openai_api_key>"

end_user = "test_user",
time_start = datetime.now(tz=timezone.UTC)
response = openai.ChatCompletion.create(
    model="gpt-3.5-turbo",
    messages=[
        {
            "role": "system",
            "content": "You are an helpful assistant"
        },
        {
            "role": "user",
            "content": "Hello, I need help with my computer"
        }
    ],
)
time_end = time_start = datetime.now(tz=timezone.UTC)
interaction = {
    "interaction": {
        "input": messages[-1]["content"],
        "output": response.choices[0].message.content,
        "history": [],
        "time_start": time_start.isoformat(),
        "time_end": time_end.isoformat(),
        "end_user": end_user,
        "model": "gpt-3.5-turbo"
    },
    "anonymize": False,
}
url = "https://backend.nebuly.com/event-ingestion/api/v1/events/interactions"
headers = {
    "Authorization": f"Bearer {NEBULY_API_KEY}",
    "Content-Type": "application/json"
}
nebuly_response = requests.request("POST", url, json=interaction, headers=headers)

Step 3: Analyze the user behavior

Please visit the tracking sections for all the API details.

Welcome

Tracking

Embedded Pages

AB Testing

Platform users

SDK