Bharathi Srinivasan
e746bf7764
adding scripts for agentcore features; jupyter notebooks moved to workshops; reorganising folders
AgentCore + Datadog LLM observability
Deploy a Strands travel agent to AgentCore runtime with traces sent to Datadog LLM observability via OTLP HTTP.
Architecture
AgentCore runtime → travel_agent.py
└── Custom OTelTracerProvider
└── OTLPSpanExporter → https://trace.agent.{DD_SITE}/v1/traces
headers: dd-api-key, dd-otlp-source=llmobs
└── Datadog LLM observability dashboard
DISABLE_ADOT_OBSERVABILITY=true bypasses the default CloudWatch ADOT pipeline.
OTEL_SEMCONV_STABILITY_OPT_IN=gen_ai_latest_experimental enables OTel v1.37+ GenAI
semantic conventions required for Datadog LLM observability views.
Prerequisites
- Python 3.10+, uv
- AWS credentials configured
- Datadog account with API key
Quick Start
pip install bedrock-agentcore boto3 python-dotenv
cp .env.example .env
# Edit .env: set DD_API_KEY (and optionally DD_SITE for non-US1 regions)
python deploy.py
python invoke.py
# View traces: https://app.datadoghq.com/llm/traces
python cleanup.py
Datadog Regions
| Region | DD_SITE |
|---|---|
| US1 (default) | datadoghq.com |
| US3 | us3.datadoghq.com |
| US5 | us5.datadoghq.com |
| EU1 | datadoghq.eu |
| AP1 | ap1.datadoghq.com |
Files
| File | Description |
|---|---|
utils/travel_agent.py |
Agent with Datadog OTel TracerProvider |
deploy.py |
Deploys to AgentCore runtime with Datadog env vars |
invoke.py |
Invokes the deployed agent |
cleanup.py |
Deletes all created AWS resources |