Skip to content

amedinat/LLMeter

BranchesTags

Repository files navigation

LLMeter

Control your AI spend across all providers — from a single dashboard.

LLMeter is an open-source cost monitoring platform for LLM APIs. Connect your provider API keys, and LLMeter automatically tracks your usage, calculates real costs, and alerts you before surprise bills hit.

Try it now: llmeter.org (free tier, no credit card required)

LLMeter Dashboard

How It Works

LLMeter connects directly to your providers' usage and billing APIs. No proxies, no code changes, no SDKs to install.

  1. Sign up at llmeter.org (or self-host)
  2. Connect your provider API keys (encrypted with AES-256-GCM at rest)
  3. LLMeter polls each provider's usage API via background cron jobs
  4. You see unified costs, trends, and alerts in one dashboard
Your App → calls OpenAI/Anthropic/etc. normally (no changes)
                         ↓
LLMeter → polls provider usage APIs → normalizes data → dashboard

Important: LLMeter reads your usage data — it does NOT intercept, proxy, or modify your API calls. Your code stays untouched.

Per-Customer Cost Attribution

Know which of your customers are costing you money. Send usage events via the ingestion API, and LLMeter breaks down costs by customer, model, and time period.

curl -X POST https://llmeter.org/api/ingest \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '[{
    "model": "gpt-4o",
    "input_tokens": 1500,
    "output_tokens": 800,
    "customer_id": "cust_abc123"
  }]'

Or use the SDK for automatic per-call attribution:

import LLMeter, { wrapOpenAI } from 'llmeter';
const llmeter = new LLMeter({ apiKey: 'lm_...' });
const openai = wrapOpenAI(new OpenAI(), llmeter, 'cust_abc123');
// every call is automatically attributed to cust_abc123

Supported Providers

Provider Integration Key type needed
OpenAI Billing API (zero-code) Admin/org key
Anthropic Billing API (zero-code) Org API key
Mistral Billing API (zero-code) Standard API key
DeepSeek Billing API (zero-code) Standard API key
OpenRouter Billing API (zero-code) — 500+ models Management API key
Google AI SDK wrapper (wrapGoogleAI) Standard API key
Azure OpenAI SDK wrapper (wrapAzureOpenAI) endpoint::apiKey format
AWS Bedrock SDK wrapper (wrapBedrock) AWS credentials

npm SDK

For providers without billing APIs (Google AI, Azure OpenAI, AWS Bedrock) or for per-request attribution, install the zero-dependency SDK:

npm install llmeter

Wrap your existing client once — every call is tracked automatically, with no changes to your base URL or request handling:

import LLMeter, { wrapOpenAI, wrapAnthropic, wrapGoogleAI, wrapAzureOpenAI, wrapBedrock } from 'llmeter';

const llmeter = new LLMeter({ apiKey: 'lm_...' });

// OpenAI
const openai = wrapOpenAI(new OpenAI(), llmeter);

// Anthropic
const anthropic = wrapAnthropic(new Anthropic(), llmeter);

// Google AI (Gemini)
const genai = wrapGoogleAI(new GoogleGenerativeAI(apiKey), llmeter);

// Azure OpenAI
const azure = wrapAzureOpenAI(new AzureOpenAI({ endpoint, apiKey }), llmeter);

// AWS Bedrock
const bedrock = wrapBedrock(new BedrockRuntimeClient({ region }), llmeter);

All wrappers: zero dependencies, ESM + CJS dual build, works in Node.js, Edge, and the browser.

Grafana / Prometheus

LLMeter exposes a Prometheus-compatible metrics endpoint that you can scrape directly into Grafana:

GET /api/v1/metrics
Authorization: Bearer YOUR_API_KEY

Returns cost_usd, requests, input_tokens, and output_tokens metrics, labeled by provider and model. Optional from / to date params for historical scraping.

# prometheus.yml
scrape_configs:
  - job_name: llmeter
    static_configs:
      - targets: ['llmeter.org']
    metrics_path: /api/v1/metrics
    bearer_token: YOUR_API_KEY

Features

  • Unified Dashboard — All providers, all models, one view
  • Per-Customer Costs — Know which users are costing you money (Team+)
  • Cost Tracking — Usage and spend pulled directly from provider billing APIs
  • Budget Alerts — Daily and monthly thresholds with email + Slack webhook notifications
  • Anomaly Detection — Statistical z-score alerting when spend deviates from your normal pattern (Pro)
  • Cost Optimization — Model swap recommendations with estimated savings
  • Usage Trends — Analyze consumption by model, provider, and time period
  • Grafana / Prometheus — GET /api/v1/metrics Prometheus endpoint, scrape directly into Grafana
  • OpenRouter Support — Track 500+ models with a single key (Pro)
  • CSV & PDF Export — Download usage reports for any date range (Pro+)
  • REST API v1 — Programmatic access to usage, providers, and customers
  • Team Management — Invite members, assign roles, shared organization billing (Team)
  • Encrypted Storage — API keys encrypted with AES-256-GCM, never stored in plain text
  • Open Source — AGPL-3.0, audit the code yourself

What LLMeter Does NOT Do

LLMeter is a monitoring and alerting tool, not a proxy or gateway. It cannot:

  • Block or throttle API calls in real-time (it reads usage after the fact)
  • Act as a circuit breaker for runaway scripts
  • Modify or intercept your requests to providers

If you need real-time request blocking, you need an AI gateway/proxy (like LiteLLM or Portkey). LLMeter complements those tools by giving you the cost visibility layer.

Pricing

Free Pro ($19/mo) Team ($49/mo)
Providers 1 Unlimited Unlimited
Budget alerts 1 Unlimited Unlimited
Data retention 30 days 1 year Unlimited
OpenRouter — ✓ ✓
Anomaly detection — ✓ ✓
Per-customer costs — — ✓
Team members — — Up to 5
Trial — 7 days free 7 days free

Self-Hosting

LLMeter can be self-hosted if you prefer to keep everything on your infrastructure.

Prerequisites

  • Node.js 20+
  • pnpm 9+
  • Supabase project (PostgreSQL + Auth)
  • Resend account (for email alerts)

Setup

git clone https://github.com/amedinat/LLMeter.git
cd LLMeter
pnpm install
cp .env.local.example .env.local
# Fill in your Supabase and other keys
# Run Supabase migrations via the Supabase dashboard or CLI
pnpm dev

Email alerts (Resend)

Budget and anomaly alerts are sent via Resend. To enable them in self-hosting or production:

  1. Create an API key at https://resend.com/api-keys → set RESEND_API_KEY.
  2. Add and verify your sending domain at https://resend.com/domains. Resend provides DNS records (SPF/DKIM) that you must add at your DNS provider (Cloudflare, Route 53, etc.). Verification typically completes in 5–15 min.
  3. Set EMAIL_FROM to an address on that verified domain (e.g. LLMeter <alerts@yourdomain.com>).

If the domain is not verified, the in-app Test button on /alerts will display the exact Resend error (e.g. Resend rejected send: The domain is not verified). For quick local testing without a domain, use LLMeter <onboarding@resend.dev> as EMAIL_FROM.

Tech Stack

  • Frontend: Next.js 16 (App Router), Tailwind CSS 4, Shadcn UI
  • Backend: Next.js API Routes, Vercel Cron (background polling)
  • Database: PostgreSQL via Supabase (with Row Level Security)
  • Auth: Supabase Auth (email/password with email verification)
  • Payments: Paddle
  • Rate Limiting: Upstash Redis (distributed, fixed-window)

Architecture

┌─────────────┐     cron jobs        ┌──────────────────┐
│ Vercel Cron  │ ──────────────────→ │  Provider APIs   │
│ (scheduled)  │                     │  (OpenAI, etc.)  │
└──────┬──────┘                      └────────┬─────────┘
       │                                      │
       │  stores normalized                   │  usage data
       │  usage records                       │
       ▼                                      ▼
┌──────────────┐                    ┌──────────────────┐
│   Supabase   │ ←───────────────── │  Adapter Layer   │
│  PostgreSQL  │    insert/upsert   │  (normalizes)    │
└──────┬───────┘                    └──────────────────┘
       │
       │  queries (RLS-protected)
       ▼
┌──────────────┐
│  Next.js 16  │ → Dashboard, Alerts, Ingest API, Prometheus
│  App Router  │
└──────────────┘

Roadmap

  • Multi-provider support (OpenAI, Anthropic, Google AI, Mistral, DeepSeek, OpenRouter)
  • Azure OpenAI + AWS Bedrock via SDK wrappers
  • Cost tracking from provider APIs
  • Budget alerts (email + Slack webhook)
  • Anomaly detection
  • Per-customer cost attribution
  • Cost optimization recommendations
  • npm SDK (llmeter) — wrappers for OpenAI, Anthropic, Google AI, Azure OpenAI, AWS Bedrock
  • Grafana/Prometheus metrics endpoint
  • CSV + PDF export
  • Billing (Paddle)
  • Fine-grained team permissions
  • Usage forecasting

Contributing

Contributions welcome! Please open an issue first to discuss what you'd like to change.

License

AGPL-3.0 — see LICENSE.

Built by John Medina | Simplifai

About

Open-source LLM cost monitoring & optimization platform. Track spend across OpenAI, Anthropic, DeepSeek & OpenRouter in real-time.

llmeter.org

Topics

open-source nextjs openai developer-tools ai-tools llm anthropic cost-tracking llm-monitoring

Resources

Readme

License

AGPL-3.0 license

Contributing

Contributing
Activity

Stars

4 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages