These notebooks introduce you to Datadog's LLM Observability Python SDK using hands-on examples.

For a detailed instrumentation guide, see Trace an LLM Application.

• A Datadog API key
• An OpenAI API key

python -m venv myenv
source myenv/bin/activate

DD_API_KEY=<YOUR_DATADOG_API_KEY>
DD_SITE=<YOUR_DATADOG_SITE>
DD_LLMOBS_AGENTLESS_ENABLED=1
DD_LLMOBS_ML_APP="onboarding-quickstart"

• Note: if your Datadog site (DD_SITE) is not provided, the value defaults to "datadoghq.com"
• Feel free to update the DD_LLMOBS_ML_APP variable to any custom app name.
• DD_LLMOBS_AGENTLESS_ENABLED=1 is only required if the Datadog Agent is not running. If the agent is running in your production environment, make sure this environment variable is unset.

OPENAI_API_KEY=<YOUR_OPENAI_API_KEY>

pip install -r requirements.txt

You can either start Jupyter on the command line (jupyter notebook) to use the web interface, or open your notebook from your preferred code editor (for example, VS Code) and run it there.

This notebook shows you how to create and trace a simple LLM call.

This notebook shows you how to create and trace a more complex, static series of steps that involves a tool call in addition to a call to an LLM.

This notebook shows you how to create and trace an LLM powered agent that calls tools and makes decisions based on data about what to do next.

This notebook shows you how to create, trace, and evaluate a RAG workflow.

When you're done with the tutorials, deactivate your virtualenv and return to your system's default Python env:

deactivate