Get started with OneAgent and AI Observability
- Latest Dynatrace
- Getting started guide
- 4-min read
- Published Jun 09, 2026
OneAgent provides automatic, zero-instrumentation observability for AI applications. With OneAgent running on the host, Dynatrace captures GenAI-specific attributes (such as the model name, token counts, latency, and prompt content) without any changes to your application code. No instrumentation libraries or SDK wrappers are required.
This guide covers Python AI SDKs for AWS Bedrock, OpenAI, and LangChain, and provides guidance for other supported providers.
Who is this for?
This getting started guide is for:
- AI engineering teams building agent- and LLM-powered applications and services who want observability without manual code changes.
- Site Reliability Engineers responsible for monitoring AI workloads.
- Platform engineers deploying Dynatrace OneAgent.
What will you learn?
By following this guide, you will learn:
- How to enable Python monitoring and the required OneAgent features for AI observability.
- How to ensure distributed trace context propagates correctly from HTTP entry-point spans into AI provider calls.
- What AI observability data OneAgent automatically captures, and how to view it in Dynatrace.
Prerequisites
In order for this to work, you need to have:
- A running Python AI application with OneAgent installed on the host.
- Dynatrace SaaS with a Dynatrace Platform Subscription (DPS) license.
- A OneAgent version that supports your AI SDK. For minimum version requirements per provider, see OneAgent platform and capability support matrix.
Prior knowledge
It's helpful to have some basic knowledge of:
- Python.
- One of the supported AI SDKs, such as boto3/botocore for AWS Bedrock, OpenAI, or LangChain.
- Configuring OneAgent features via
Settings.
Get started with OneAgent and AI observability
1. Enable Python monitoring
Enable Python monitoring for your OneAgent as described in Python. This is required for OneAgent to instrument your Python AI application.
2. Enable required OneAgent features
Navigate to Settings > Collect and capture > General monitoring settings > OneAgent features and filter by Python.
Each supported AI SDK has a dedicated OneAgent feature that must be enabled to activate monitoring for that provider.
Enable the feature that matches your AI SDK, plus any framework and optional features relevant to your application:
| Feature | Purpose | Required |
|---|---|---|
Python AWS SDK Client | Instruments botocore; gates all AWS SDK monitoring | Required |
Python AWS SDK GenAI Bedrock | Adds Bedrock-specific GenAI attributes (model, tokens, …) | Required |
Python AWS SDK Bedrock prompt capture | Captures prompt text and model responses | Optional |
Python GenAI Langchain | Adds LangChain chain/node spans alongside Bedrock spans | Optional |
If your application serves HTTP requests, also enable the OneAgent feature for your web framework (for example, Python FastAPI, Python Django, or Python Flask) to create HTTP entry-point spans that the AI provider spans can nest under.
Sensors for additional providers are available as experimental sensors. These include, for example, Anthropic, Cohere, Groq, Mistral AI, Ollama, Google GenAI, CrewAI, and Haystack.
Experimental sensors are functional but delivered on an accelerated schedule to extend GenAI observability to additional AI providers.
They are not covered by Dynatrace support SLAs, and attribute collection is best-effort; completeness and schema may change in future releases without prior notice.
To enable an experimental sensor, search for the name of your AI provider or SDK in Settings > Collect and capture > General monitoring settings > OneAgent features and enable the corresponding feature flag.
Prompt capture is currently supported for AWS Bedrock and OpenAI, including when those providers are used as the underlying model with LangChain.
Enable the corresponding optional prompt capture feature to view prompt content in 
AI Observability > Explorer.
After enabling the OneAgent features, restart the Python application so the new features are picked up at process start.
3. Deploy and send test requests
Start your application and verify that OneAgent is monitoring the process.
You can confirm this by checking that the process is visible in 
Infrastructure & Operations > Explorer > Compute > Processes in the Dynatrace UI.
Send a test request to your application:
curl -X POST http://<your-service>/generate \-H 'Content-Type: application/json' \-d '{"prompt":"What is observability?"}'
Expected response:
{"result": "..."}
4. Observe in Dynatrace
Navigate to AI Observability > Explorer.
The application appears as an AI service once the first instrumented request completes.
LLM request count and token usage are aggregated and visible at the service level.
Select the service to open the detail view. The Prompts chart shows latency data for captured LLM requests.
To inspect the full distributed trace for an individual request, navigate to 
Distributed Tracing.
The HTTP entry-point span (from the web framework sensor) contains the AI provider call as a child span carrying all GenAI-specific attributes:
gen_ai.provider: The name of the AI provider (for example,aws_bedrockoropenai).gen_ai.model: The model used for the request.gen_ai.usage.input_tokens/gen_ai.usage.output_tokens: Token counts for the request.
Congratulations!
Now that you've set up OneAgent to automatically observe your AI application, you can:
- Explore the AI Observability app to visualize your AI workloads.
- Navigate to Distributed Tracing to inspect individual AI provider spans.
- Check out the sample applications for more examples.
AI Observability