@aivyuh/finops

Transparent cost telemetry for Anthropic and OpenAI Node.js clients. Part of the V7 AI FinOps platform.

Wrap your existing AI SDK client in one line — @aivyuh/finops captures model, tokens, cost, and latency metadata and sends it to the V7 backend. It never reads or stores prompt or response content.

Install

npm install @aivyuh/finops

Quickstart

Anthropic

import Anthropic from "@anthropic-ai/sdk";
import { wrapClient } from "@aivyuh/finops";

const client = wrapClient(new Anthropic(), {
  telemetryEndpoint: "https://finops-api.aivyuh.com/telemetry",
  customerId: "cust-123",
  apiKey: process.env.AIVYUH_FINOPS_KEY, // per-customer key, sent as X-API-Key
  project: "my-app",
  tags: { feature: "chat", team: "product" },
});

// Use the client exactly as before — all types and overloads are preserved
const message = await client.messages.create({
  model: "claude-sonnet-4-6-20260320",
  max_tokens: 1024,
  messages: [{ role: "user", content: "Hello, Claude!" }],
});

OpenAI

import OpenAI from "openai";
import { wrapClient } from "@aivyuh/finops";

const client = wrapClient(new OpenAI(), {
  telemetryEndpoint: "https://finops-api.aivyuh.com/telemetry",
  customerId: "cust-123",
  project: "my-app",
  tags: { feature: "search", team: "platform" },
});

// Use the client exactly as before
const completion = await client.chat.completions.create({
  model: "gpt-4o",
  messages: [{ role: "user", content: "Hello, GPT!" }],
});

Configuration

| Option | Type | Required | Description | |--------|------|----------|-------------| | telemetryEndpoint | string | Yes | V7 telemetry ingest URL (https://finops-api.aivyuh.com/telemetry) | | customerId | string | Yes | Your customer ID for cost attribution | | apiKey | string | Yes* | Per-customer API key, sent as X-API-Key header. Required by the production V7 ingest endpoint. | | project | string | No | Project name for grouping costs | | tags | Record<string, string> | No | Custom key-value tags attached to every request | | disableTelemetry | boolean | No | Set true to disable (useful in tests) |

Production API Endpoint

https://finops-api.aivyuh.com/telemetry

POST telemetry payloads are accepted with a 202 Accepted response. The endpoint is fire-and-forget — if unreachable, your application is unaffected.

How It Works

1. wrapClient() detects whether you passed an Anthropic or OpenAI client (duck-typing).
2. It monkey-patches messages.create() (Anthropic) or chat.completions.create() (OpenAI).
3. After each successful call, it extracts usage metadata (model, tokens, latency) and estimates cost.
4. A non-blocking fetch() sends the telemetry payload to V7 — errors are silently swallowed.
5. Your original response is returned untouched.

Advanced: Individual Wrappers

If you only use one provider, import the specific wrapper to avoid unused type imports:

import { wrapAnthropic } from "@aivyuh/finops";
import { wrapOpenAI } from "@aivyuh/finops";

Requirements

• Node.js >= 18 (uses native fetch)
• @anthropic-ai/sdk >= 0.40.0 and/or openai >= 1.50.0 as peer dependencies

License

MIT