agentmetrics 0.1.2

Monitor your AI agents. Track cost, latency, and errors in two lines of code.

agentmetrics — Python SDK

Production observability for AI agents. Track cost, latency, tokens, failures, and tool calls — one decorator, zero infrastructure.

→ agentmetrics.dev/docs/sdk/python

---

Install

pip install agentmetrics

Requires Python 3.9 or later.

---

Quickstart

import os
import agentmetrics

agentmetrics.configure(api_key=os.environ["AGENTMETRICS_API_KEY"])
agentmetrics.instrument()  # auto-patches OpenAI, Anthropic, LiteLLM, and more

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

Every call to run() reports to your dashboard: duration, cost, token usage, success/failure, and error messages.

---

API

agentmetrics.configure()

Call once at startup before any @track decorators execute.

agentmetrics.configure(
    api_key="am_live_...",       # required
    environment="production",    # optional — tags every run
    sample_rate=1.0,             # optional — float 0.0–1.0
    base_url="https://...",      # optional — override for self-hosted
    batch_size=20,               # optional — events per batch
    flush_interval=2.0,          # optional — seconds between flushes
)

agentmetrics.instrument()

Patches installed LLM SDKs to auto-capture token counts and model names on every call. Idempotent.

agentmetrics.instrument()

Supported: OpenAI (+ Azure, Groq, Together AI) · Anthropic · LiteLLM · Google Gemini · Cohere · Mistral · LangChain / LangGraph / CrewAI · LlamaIndex

@agentmetrics.track()

Decorator for sync and async agent functions.

@agentmetrics.track(agent_id="my-agent", metadata={"env": "prod"})
def run(task: str) -> str:
    return call_llm(task)

@agentmetrics.track(agent_id="async-agent")
async def run_async(task: str) -> str:
    return await call_llm_async(task)

Parameter	Type	Description
agent_id	str	Identifier shown in the dashboard
metadata	dict	Optional key-value pairs attached to every run

agentmetrics.step()

Context manager for timing a named phase within a tracked agent.

@agentmetrics.track(agent_id="pipeline")
def run(query: str) -> str:
    with agentmetrics.step("retrieve"):
        docs = vector_search(query)
    with agentmetrics.step("generate"):
        return call_llm(query, docs)

agentmetrics.tool()

Context manager for tracking individual tool calls.

@agentmetrics.track(agent_id="research-agent")
def run(query: str) -> str:
    with agentmetrics.tool("web_search"):
        results = web_search(query)
    return summarize(results)

agentmetrics.flush()

Blocks until all queued events are sent. Call before process exit in scripts.

agentmetrics.flush(timeout=10.0)

agentmetrics.trace_id

Returns the active trace ID from inside a tracked function.

@agentmetrics.track(agent_id="my-agent")
def run(task: str) -> str:
    print(agentmetrics.trace_id)  # e.g. "a3f1c2d4-..."
    return call_llm(task)

---

License

MIT