Skip to main content
Get started with our SDK and ensure reliability for your app with intelligent observability and evaluation. Follow this quick guide to plug LLUMO AI into your workflow—fast, simple, and built for AI product teams.

​
1. Install the LLUMO AI SDK

Please ensure that you have the LLUMO AI SDK installed before running evaluation. 
!pip install llumo

​
2. Set your API key

You will find your LLUMO AI API key on the top navbar next to your profile icon. 
LLUMO_AI_API_KEY = "key_NzE4NmU4*****************00d0cd"

​
3. Select your framework

Simply select your framework. For different frameworks, we have different callback handlers or wrappers. If you just want to evaluate the final log, select evaluate 
!pip install langchain_community
!pip install langchain-openai

from langchain_openai import ChatOpenAI
from langchain.agents import initialize_agent, Tool, AgentType
from llumo import LlumoLogger

logger = LlumoLogger(apiKey=llumoKey, playground="playground-agent-logs")

sessionContext = LlumoSessionContext(logger=logger)
sessionContext.start()

sessionContext.startLlumoRun(runName="QUERY_EXECUTION")
callbackHandler = LlumoCallbackHandler(session=sessionContext)

try:
    # Execute query
    result = agent.run(q, callbacks=[callbackHandler])
except Exception as e:
    print(f"❌ Error: {str(e)}")

# End Run
sessionContext.endLlumoRun()

# End Session
sessionContext.end()

from llumo.openai import OpenAI
from llumo.llumoLogger import LlumoLogger
from llumo.llumoSessionContext import LlumoSessionContext

logger = LlumoLogger(apiKey=llumoKey, playground="playground-openai-raw")

# Initialize and start the Llumo Session
session = LlumoSessionContext(logger = logger)


client = OpenAI(
    api_key="openaiKey",
    session=session,
)

# Start Session
session.start()
session.startLlumoRun(runName="QUERY_EXECUTION")

messages = [{"role": "user", "content": query}]

# Output Generation
response = client.chat.completions.create(
    model="gpt-3.5-turbo",
    messages=messages,
    createExperiment=True,
)

aiOutput = response.choices[0].message.content
results.append({"query": query, "output": aiOutput})

# End Run
session.endLlumoRun()

# End Session
session.end()



from llumo import LlumoLogger, LlumoSessionContext

# Initialize logger
logger = LlumoLogger(
    apiKey=llumoKey,
    playground="playground-batch-eval"
)

# Initialize Llumo session
session = LlumoSessionContext(logger=logger)

# Run multiple evaluations
resData = session.evaluateMultiple(
    data,
    evals=[
        "Response Correctness",
        "Hallucination",
        "Response Completeness",
        "Input Bias",
    ],
    createExperiment=True,  # Creates a named experiment in your dashboard
)

# Print results
print(resData)

​
4. View Results on the LLUMO AI Dashboard

Once you have integrated the SDK, we will handle all the heavy lifting, from logging, evaluating, creating run-reports, analysis issues across different runs and creating flow graph and insights for you. You can check out the whole debug reports on app.llumo.ai/debugger
⌘I