Quickstart

Server running, first agent tracked, data in the dashboard — in under five minutes.

1. Start the AgentMetrics server

AgentMetrics runs as a local server. Your SDK sends data to it. Start the server before doing anything else.

pip install agentmetrics-server
agentmetrics install

Dashboard: http://localhost:3099 API: http://localhost:8099

2. Get an API key

Open http://localhost:3099 and go to Settings to generate a key.

3. Install the SDK

pip install agentmetrics

4. Configure

Call configure() once at startup, before any tracking calls.

import os
import agentmetrics

agentmetrics.configure(
    api_key=os.environ["AGENTMETRICS_API_KEY"],
    base_url="http://localhost:8099",
)

5. Enable token and cost tracking

Call instrument() once after configure(). It patches LLM client libraries to capture token counts and cost on every call.

agentmetrics.instrument()

6. Wrap your agent

@agentmetrics.track(agent_id="my-agent")
def my_agent(task: str) -> str:
    answer = call_llm(task)
    return answer

7. Run it and check the dashboard

result = my_agent("summarize this document")

Open http://localhost:3099 — your run appears within a few seconds.

What gets tracked automatically

Every invocation captures:

• Duration (wall-clock, from call to return)
• Success or failure status
• Error message if the function raises or throws
• Cost and token usage across all LLM calls (when instrument() is called)

Next steps

Using a framework instead of the raw SDK?

• LangChain: Python and JavaScript
• CrewAI: Python
• LlamaIndex: Python
• OpenAI Agents SDK: Python
• AutoGen: Python
• Anthropic Managed Agents: Python and JavaScript
• Hermes plugin: zero code
• OpenClaw plugin: zero code

Full SDK reference: Python SDK · JavaScript SDK

Deploying for a team or to the cloud? See Deploy.