opsmeter-io-sdk 0.1.0b2

Official opsmeter.io Python SDK: one-time init + request context auto-instrumentation

opsmeter-io-sdk (Official opsmeter.io SDK)

Python SDK preview for Opsmeter auto-instrumentation. PyPI package: opsmeter-io-sdk Integration examples: opsmeter-integration-examples Opsmeter site: https://opsmeter.io Official publisher identity: opsmeter.io.

Use this SDK for LLM cost tracking, OpenAI usage monitoring, Anthropic usage telemetry, and no-proxy AI observability in Python.

Provider/model names should come from: https://opsmeter.io/docs/catalog Current SDK provider support: OpenAI and Anthropic only.

Quick links

• Product: https://opsmeter.io
• Docs: https://opsmeter.io/docs
• Model catalog: https://opsmeter.io/docs/catalog
• Integration examples: https://github.com/opsmeter-io/opsmeter.io-integration-examples

Model catalog (required)

Always use provider/model pairs from the official catalog: https://opsmeter.io/docs/catalog

Examples:

• OpenAI: provider=openai, model=gpt-4o-mini
• Anthropic: provider=anthropic, model=claude-3-5-sonnet-20241022

Install

pip install opsmeter-io-sdk openai
# optional:
# pip install anthropic

Core model

• init(...) once at process startup (idempotent)
• request-level attribution via context(...)
• provider call stays direct (no proxy)
• telemetry emit is async and non-blocking by default

Telemetry usage (no options)

import opsmeter_sdk as opsmeter
from openai import OpenAI

opsmeter.init(
    api_key="...",
    environment="prod",
)

client = OpenAI()

opsmeter.capture_openai_chat_completion(
    lambda: client.chat.completions.create(
        model="gpt-4o-mini",
        messages=[{"role": "user", "content": "hello"}],
    )
)

Telemetry usage (with options/context)

import opsmeter_sdk as opsmeter
from openai import OpenAI
from anthropic import Anthropic

opsmeter.init(
    api_key="...",
    workspace_id="ws_123",
    environment="prod",
)

openai = OpenAI()
anthropic = Anthropic()

with opsmeter.context(
    user_id="u_1",
    tenant_id="tenant_a",
    endpoint="/api/chat",
    feature="assistant",
    prompt_version="v12",
    data_mode="real",
):
    openai_captured = opsmeter.capture_openai_chat_completion_with_result(
        lambda: openai.chat.completions.create(
            model="gpt-4o-mini",
            messages=[{"role": "user", "content": "hello"}],
        ),
        request={"model": "gpt-4o-mini"},
        await_telemetry_response=True,
    )

with opsmeter.context(
    user_id="u_1",
    tenant_id="tenant_a",
    endpoint="/api/support",
    feature="support",
    prompt_version="v8",
):
    anthropic_captured = opsmeter.capture_anthropic_message_with_result(
        # Provider/model names: https://opsmeter.io/docs/catalog
        lambda: anthropic.messages.create(
            model="claude-3-5-sonnet-20241022",
            max_tokens=128,
            messages=[{"role": "user", "content": "Summarize this support ticket."}],
        ),
        request={"model": "claude-3-5-sonnet-20241022"},
        await_telemetry_response=True,
    )

print(openai_captured["telemetry"])     # { ok, status, body? }
print(anthropic_captured["telemetry"])  # { ok, status, body? }

API

• init(...)
• context(...)
• get_context()
• capture_openai_chat_completion(...)
• capture_openai_chat_completion_with_result(...)
• capture_openai_chat_completion_async(...)
• capture_anthropic_message(...)
• capture_anthropic_message_with_result(...)
• capture_anthropic_message_async(...)
• patch_openai_client(...)
• flush()

Tests

python3 -m py_compile opsmeter_sdk/sdk.py tests/test_sdk.py
python3 -m unittest tests/test_sdk.py -v