Works with every major agent framework

Install the plugin from the OpenClaw registry. No changes to your agent code needed. Every run, tool call, and cost tracked automatically.

openclaw plugins install agentmetrics-openclaw

Add the @agentmetrics.track() decorator to your Hermes agent class or function. Captures agent runs, tool calls, memory reads, token usage, and cost.

pip install agentmetrics

# Then decorate your agent:
@agentmetrics.track()
class MyAgent(Agent): ...

Wrap your LangChain agent or chain function with the @agentmetrics.track() decorator. Captures the full run including all LLM calls, tool calls, and the final result.

pip install agentmetrics

@agentmetrics.track()
def run_chain(input: str) -> str:
    return chain.invoke({"input": input})

Wrap your CrewAI Crew with the AgentMetrics decorator to capture each crew run as a structured event, including all agent and task executions within the crew.

pip install agentmetrics

@agentmetrics.track()
def run_crew(input: str) -> str:
    return my_crew.kickoff(inputs={"input": input})

Wrap your LlamaIndex query engine or agent function with the @agentmetrics.track() decorator. Captures the full run including retrieval steps, synthesis, and LLM calls.

pip install agentmetrics

@agentmetrics.track()
def run_query(query: str) -> str:
    return query_engine.query(query)

Instrument AutoGen conversations and agent groupchats. Each conversation thread is captured as a run, with individual agent messages tracked as events.

pip install agentmetrics

@agentmetrics.track()
def run_groupchat(task: str):
    user_proxy.initiate_chat(assistant, message=task)

Track OpenAI Agents SDK runs including all tool calls and handoffs. Works with single agents and multi-agent handoff patterns.

pip install agentmetrics

@agentmetrics.track()
async def run_agent(input: str):
    result = await Runner.run(my_agent, input)
    return result.final_output

AgentMetrics works with any Python code. Decorate any function that represents a unit of agent work and the SDK will capture its execution automatically.

pip install agentmetrics

@agentmetrics.track()
def my_agent(task: str) -> str:
    # Any Python code: LLM calls, tool calls, loops
    # AgentMetrics captures it all
    ...