fluxmeter 1.1.0

Python SDK for FluxMeter — streaming metering for AI token billing

FluxMeter Python SDK

Send AI token usage events to FluxMeter for real-time aggregation and billing.

Install

pip install fluxmeter

Quick Start — Lite mode (default, no Kafka)

from fluxmeter import FluxMeter

meter = FluxMeter(api_url="http://localhost:8000")  # default
meter.track("cust_123", "gpt-4o", input_tokens=500, output_tokens=150)

Runs against make demo (API → Redis). Lite /ingest returns cost_usd and balance_usd in the response.

Full stack (Kafka → Flink)

meter = FluxMeter(kafka_brokers="localhost:9094", wal_enabled=True)
meter.track("cust_123", "gpt-4o", input_tokens=500, output_tokens=150)

OpenAI Integration

import time
from openai import OpenAI
from fluxmeter import FluxMeter

client = OpenAI()
meter = FluxMeter(api_url="http://localhost:8000", environment="production")

start = time.time()
response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Hello!"}],
)
latency = int((time.time() - start) * 1000)

meter.track_openai("cust_123", response, latency_ms=latency)

Anthropic Integration

import anthropic
from fluxmeter import FluxMeter

client = anthropic.Anthropic()
meter = FluxMeter(api_url="http://localhost:8000")

response = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello!"}],
)

meter.track_anthropic("cust_123", response)

Multi-tenant (SaaS)

meter.track(
    customer_id="cust_123",
    model_id="gpt-4o",
    tenant_id="tenant_acme",
    input_tokens=100,
    output_tokens=50,
)

Use with Flink full mode and tenantId on events for Redis key isolation.

Manual Tracking (any provider)

meter.track(
    customer_id="cust_123",
    model_id="gemini-1.5-pro",
    provider="google",
    input_tokens=2000,
    output_tokens=500,
    request_id="req_abc123",
    span_id="span_7f3a",
    session_id="sess_456",
    latency_ms=890,
    environment="production",
    metadata={"feature": "code-review", "team": "platform"},
)

Configuration

# HTTP (lite) with API key
meter = FluxMeter(
    api_url="https://meter.example.com",
    api_key="fm_live_...",
)

# Kafka (full) with WAL and SASL
meter = FluxMeter(
    kafka_brokers="kafka1:9092,kafka2:9092",
    topic="token-events",
    environment="production",
    wal_enabled=True,
    wal_path="~/.fluxmeter/wal",
    producer_config={
        "security.protocol": "SASL_SSL",
        "sasl.mechanisms": "PLAIN",
        "sasl.username": "...",
        "sasl.password": "...",
    },
)

Parameter	Default	Description
api_url	http://localhost:8000	HTTP ingest base URL (used when kafka_brokers is omitted)
kafka_brokers	None	If set, sends to Kafka instead of HTTP
api_key	None	X-API-Key header for HTTP ingest

How It Works

Lite:

Your App  →  meter.track(...)  →  POST /ingest  →  Redis (atomic Lua)

Full:

Your App  →  meter.track(...)  →  Kafka  →  Flink  →  Redis  →  API / Grafana

Kafka mode batches with lz4 compression and optional WAL for zero data loss during broker outages.

Requirements

• Python 3.9+
• confluent-kafka (required for Kafka mode; installed with package)
• FluxMeter API running (make demo) or full stack (make demo-full)

Release

See docs/pypi-release.md.