Overview

MetricAI is financial rails for your AI agents. Route your LLM, voice, and tool calls through the MetricAI proxy. Meter usage cost in INR and emit telemetry - without changing your provider code.

PyPI: pip install metricai
BYOK - bring your own keys
Fail-open - telemetry never blocks agents
INR - wallet-native billing

Copy-paste quickstart

Install with the Installation guide. Full walkthrough in Quickstart.

python

import os
import metricai
from openai import OpenAI
from dotenv import load_dotenv

load_dotenv()

metricai.init(
    api_key=os.getenv("METRICAI_API_KEY"),
    auto_instrument=True,
    default_agent_id="weather-bot",
    default_user_id="user_123",
    llm_keys={"openai": os.getenv("OPENAI_API_KEY")},
)

client = OpenAI()
stream = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "What is the current weather in Tokyo?"}],
    stream=True,
)
for chunk in stream:
    if chunk.choices and chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="", flush=True)

Core concepts

Concept	What it is
`Proxy routing`	Calls go to …/v1/proxy/{provider} for server-side metering.
`BYOK`	You keep provider keys; MetricAI forwards via X-*-API-Key headers. Default (mode="byok").
`Managed`	MetricAI-managed provider keys; opt in via mode="managed".
`Attribution`	agent_id, user_id, session_id, workflow ids for cost breakdowns.
`Telemetry / track()`	Events POST to /v1/sdk/events. Fail-open by default.
`Sessions`	Group turns under one session_id with aggregates and budget caps.

Fail-open guarantee: telemetry and metering errors are logged and never raise unless raise_on_error=True.

Supported providers

Provider	Support
`OpenAI`	Streaming chat, Responses, web search
`Anthropic`	Streaming Messages + web search
`Gemini`	Streaming Interactions + generateContent
`Grok`	Streaming xAI gRPC SDK
`Groq`	Streaming LPU chat completions
`Perplexity`	Streaming Sonar + Search API
`Vertex AI`	Streaming google-genai with GCP SA
`Azure OpenAI / Foundry`	Streaming OpenAI SDK + azure llm_keys
`Mistral`	Streaming LangChain ChatMistralAI
`Bedrock`	converse_stream + invoke_model
`Tavily`	search, extract, crawl, map, qna, research
`Voice`	Deepgram, ElevenLabs, Sarvam
`LangChain / LangGraph`	Native chat models + stream()

Next Quickstart

MetricAI SDK · PythonView on PyPI