Track LLM costs, token usage, and model spend in one dashboard. LLMetrics shows what each feature costs, surfaces usage spikes in real time, and helps you fix runaway AI spend before it hits your invoice.
SDK setup in under 5 minutes · Free plan includes 10,000 events/mo
Works with every major provider
The problem
You ship AI features fast. But without visibility, costs compound — and by the time the invoice arrives, tracing the cause is nearly impossible.
A single bad prompt or runaway loop can 10x your daily spend. You find out when the invoice arrives.
Engineers swap models without realizing the cost difference. GPT-4o sneaks into paths meant for GPT-4o-mini.
Your AI spend is one line item. You can't tell if it's your summarizer, your chatbot, or your search feature causing it.
Features
From a real-time LLM usage dashboard to proactive cost alerts, LLMetrics helps you understand and control AI spend.
Tag every LLM call with a feature name. LLMetrics breaks down spend, token count, and model usage per feature — so you can see if your summarizer is 10x more expensive than your chatbot.
Set spend thresholds and spike detection rules. LLMetrics fires alerts via email or Slack the moment your usage pattern looks wrong — not a month later on your invoice.
See cost per model side by side across all your features. Identify where an expensive model is being used unnecessarily and quantify exactly how much switching would save.
See spend update as events come in.
Visualize spend over time with full history.
Separate dashboards per product or team.
Invite teammates on the Team plan.
How it works
No wrappers, no proxies. Drop two calls into your existing LLM code and start seeing spend in a live dashboard.
One npm install and you're set. Works with any Node.js or edge runtime — no framework required.
Call metrics.track() after each LLM response with token counts and a feature tag. Fire-and-forget — it won't slow your app.
Open your dashboard and watch spend, tokens, and feature breakdowns update in real time.
Developer-first
No agents, no wrappers, no request proxying. Drop it into your existing LLM calls.
import { llmetrics } from "@llmetrics/sdk";
llmetrics.init({
apiKey: process.env.LLMETRICS_API_KEY,
});
// Call your LLM as normal...
const response = await openai.chat.completions.create({ ... });
// Then track it. Fire-and-forget.
llmetrics.track({
feature: "lesson-generation",
provider: "openai",
model: "gpt-4o-mini",
inputTokens: response.usage.prompt_tokens,
outputTokens: response.usage.completion_tokens,
});Supports 100+ models across OpenAI, Anthropic, Google, Mistral, and more — pricing synced daily.
Use cases
Whether you need LLM spend tracking for one product or a shared dashboard across providers, LLMetrics gives you a simple way to monitor AI cost without rebuilding your stack.
Watch daily spend trends and see which releases or prompt changes moved cost.
Track events, input tokens, output tokens, providers, and model mix in one place.
See which workflows create the biggest token load and which models cost the most.
Catch regressions with alerts and compare high-cost features before they become margin problems.
Pricing
Start free. No credit card required. Upgrade when you need more events or retention.
For indie developers and side projects.
For production AI apps that need full observability.
For teams building serious AI products.
FAQ
No. llmetrics.track() is fire-and-forget — it sends data asynchronously in the background and does not block your response path.
Install the SDK, send token counts plus provider, model, and feature metadata after each call, and LLMetrics turns that into one dashboard across OpenAI, Anthropic, Google Gemini, and more.
No. Only usage metadata is sent — token counts, model name, provider, and your feature tag. Your actual prompt and response content never leaves your server.
One event = one LLM API call tracked via llmetrics.track(). If you call GPT-4o-mini once, that's one event.
OpenAI, Anthropic, Google Gemini, Mistral, Cohere, Together AI, Groq, and more — 100+ models total with pricing synced daily. If a model isn't priced yet, the event is still recorded and costs are backfilled when pricing is added.
Yes. Cancel from your billing settings at any time. Your plan stays active through the end of the billing period.
Free to start. No credit card required. SDK setup in under 5 minutes.
Get started for free →10,000 events/month free · No credit card required