agentmetrics-langchain

AgentMetrics integration for LangChain. Pass one callback to any chain or agent .invoke() call and every run reports back to your dashboard showing latency, cost, token counts, tool calls, and errors, with no changes to your agent logic.

Install

pip install agentmetrics-langchain

Quickstart

from agentmetrics_langchain import AgentMetricsCallback

cb = AgentMetricsCallback(
    agent_id="my-langchain-agent",
    base_url="http://localhost:8099",
)

result = agent.invoke(
    {"input": "What is the weather in Paris?"},
    config={"callbacks": [cb]},
)

cb.flush()

API

AgentMetricsCallback(agent_id, base_url)

The callback is a BaseCallbackHandler. Pass it via config={"callbacks": [cb]} on any chain or agent .invoke() call. It tracks the top-level chain only, with nested sub-chains aggregated into the same run.

Supports both OpenAI-style and Anthropic-style token counting from usage_metadata and llm_output.

.flush(timeout=10.0)

Blocks until all in-flight HTTP requests complete. Call before process exit in scripts.

What gets tracked

Each top-level chain invocation emits one event to /v1/events on completion or error:

LangGraph

The callback works with LangGraph graphs the same way:

from langgraph.graph import StateGraph
from agentmetrics_langchain import AgentMetricsCallback

cb  = AgentMetricsCallback(base_url="http://localhost:8099")
app = build_graph().compile()

result = app.invoke(state, config={"callbacks": [cb]})

License

MIT