Observe Your First Agent
This tutorial walks you through registering an externally-hosted agent with WSO2 Agent Manager, connecting it with zero-code instrumentation, and viewing its traces in the AMP Console.
Step 1: Open or Create a Project​
WSO2 Agent Manager organises agents inside Projects.
- Open the AMP Console.
- Click the project switcher in the top navigation bar.
- Select an existing project (e.g. Default Project) or click + Create a Project, enter a name and description, then click Create.
Step 2: Add a New Agent​
- Inside your project, click + Add Agent.
- On the Add a New Agent screen, choose Externally-Hosted Agent — use this when your agent runs outside the AMP platform (locally, on your own infrastructure, or in a third-party cloud).
Step 3: Register the Agent​
- Fill in the Agent Details:
- Name — e.g.
my-first-agent - Description (optional)
- Name — e.g.
- Click Register.
You are taken to the agent's Overview page. A Setup Agent panel opens automatically on the right.
Step 4: Set Up Instrumentation (Setup Agent Panel)​
The Zero-code Instrumentation Guide in the Setup Agent panel walks you through four steps.
4.1 Install AMP Instrumentation Package​
pip install amp-instrumentation
4.2 Generate an API Key​
In the panel, select a Token Duration and click Generate. Copy the key immediately — it will not be shown again.
4.3 Set Environment Variables​
export AMP_OTEL_ENDPOINT="http://localhost:22893/otel"
export AMP_AGENT_API_KEY="<your-generated-api-key>"
Replace the endpoint with your AMP instance's OTLP endpoint if it differs from the default shown in the panel.
4.4 Run Your Agent with Instrumentation​
Prefix your normal run command with amp-instrument:
# Python script
amp-instrument python my_agent.py
# FastAPI / async service
amp-instrument uvicorn app:main --reload
# Poetry / uv managed projects
amp-instrument poetry run python agent.py
amp-instrument uv run python agent.py
No changes to your agent code are required.
Activate it before running amp-instrument so the package is on your path.
Step 5: View Traces in the Console​
Once your agent has handled a few requests:
- In the left sidebar, under OBSERVABILITY, click Traces.
- Each trace represents one end-to-end agent invocation. Click any trace to expand it:
- The root span shows end-to-end latency and the agent name.
- LLM spans show the model, token counts, and call latency.
- Tool spans show tool name, input, and output.