LLMtrack

Resources / Integrations

INTEGRATIONS · FOR TEAMS ALREADY ON A GATEWAY

Your gateway shows the bill. LLMtrack shows which feature and which customer ran it up.

You already route through a gateway. Paste one URL into its trace export and get cost per feature, cost per customer against what they pay you, and next month's bill. No SDK, no code change, no second proxy.

Connect your gateway free →See what you'd get

Free plan · 3,000 requests a month · no card · about 2 minutes to set up

Works withOpenRouterany OpenTelemetry exporter

SAMPLE DATA · SAME 30 DAYS, TWO VIEWS

WHAT YOUR GATEWAY SHOWS

gpt-5.6-sol$612.40

claude-sonnet-5$188.10

gemini-3.6-flash$21.70

Total$822.20

Accurate, and it doesn't tell you what to change.

THE SAME TRAFFIC IN LLMTRACK

FEATURE · SHARE · COST

code70%$573.10

tts19%$152.40

image-generation10%$82.66

summarizer2%$14.04

acme-corp pays you $29 and cost $61.20 in AI this month.

At this pace you pass your $900 budget on day 27.

HOW IT WORKS

Your gateway already writes a report for every request. Send a copy to us.

01

Your app calls the gateway

Nothing changes in your request path. We never see the request while it's happening.

02

The gateway exports a trace

After the response returns, it sends a standard OpenTelemetry record: model, token counts, cost, your labels.

03

We price it and break it down

Every component is priced against the model's published rates, then grouped by feature and customer in under a second.

WHAT YOU GET THAT YOUR GATEWAY DOESN'T SHOW

Keep your gateway. Add the business view.

Your gateway routes, retries and caps spend. These six views answer what it can't: what to change, and whether a customer is worth what they cost.

COST PER FEATURE

See which part of your product spends the money

Label each request once and every chart splits by feature. Tokens alone mislead: the cheapest feature here uses the most of them.

summarizer · 129M tokens$14.04

code · 29.7M tokens$573.10

UNIT ECONOMICS

Cost per customer, against what they pay

Pass the customer ID your gateway already accepts. We show margin per account, so you find the customers you're losing money on.

acme-corp · Pro $29−111% margin

northwind · Pro $29+89% margin

FORECAST

Next month's bill, and the day you cross budget

Projected from your real daily run rate, with the feature that's driving the growth named.

Projected close$1,012

$900 budgetpassed on day 27

Driverimage-generation +41%

CACHE INTELLIGENCE

Know the hour your caching broke

Hit rate per feature and what a collapse is costing you. Someone edits a prompt, the prefix changes, and you hear about it the same day.

docs-search · today 14:2085% → 4%

Cost of the drop+$8.40 per day

MODEL INTELLIGENCE

Switch advice that will say "don't"

Your gateway makes switching models one line. We check each switch against independent benchmarks for that feature's kind of work first.

code-review → cheaper modelsaves $43.69/mo

Coding index 73 → 52Keep

GATEWAY FEES

What you pay vs the model's list price

We show the cost your gateway reported next to our own price from the provider's published rates. The gap is your gateway's fee.

Gateway reported$0.0151

List price$0.0144

Difference$0.0007 (5%)

All figures on this page are sample data.

CONNECT IN THREE STEPS

Pick your gateway

OpenRouter

  1. 01

    Copy your LLMtrack ingest key

    Sign up free, open Keys & Integrations, and create a key named "OpenRouter". One key per connection lets you revoke it on its own.

  2. 02

    Add a Webhook in Settings › Observability

    In OpenRouter, open Settings › Observability, add a Broadcast destination of type Webhook, and paste:

    URL      https://otlp.llm-track.com/v1/traces
Headers  {"Authorization": "Bearer llmt_live_…"}

    Turn on Privacy Mode for this destination. LLMtrack drops prompt text on arrival anyway, and with Privacy Mode it's never sent.

  3. 03

    Label your requests

    Add the customer and the feature to the requests you already send:

    "user":  "acme-corp",
"trace": { "feature": "summarizer" }

    Tip Some calls also go through the LLMtrack SDK? Add responseId: completion.id to those track() calls. Each request then lands once, whichever way it arrives, and your event count stays exact.

Full OpenRouter guide, with troubleshooting →

That's it. Send a request and it shows up on your dashboard in under a second, already priced. Not on a gateway? Add the SDK in one line

Tip Running the SDK on some calls and a gateway on others, or both on the same calls? Send responseId from the SDK and each request lands once, however it reaches us.

WHAT WE KEEP, WHAT WE DROP

Gateways can send prompt text. We delete it before anything is saved.

Kept

  • model and provider
  • token counts: input, output, cached, cache writes, reasoning, audio
  • the cost your gateway reported
  • your feature label and customer ID
  • latency and status

Dropped on arrival

  • prompts and system instructions
  • completions and tool calls
  • images and audio
  • anything else that holds content

IS THIS FOR YOU

Where it helps, and where your gateway is enough

Connect it if

  • more than one feature in your product calls a model
  • you charge customers a plan price and want to know who's profitable
  • a bill has surprised you before
  • you're about to switch models to save money

Skip it if

  • you have one model and one feature. Your gateway dashboard covers that.
  • you need to read prompts and debug chains. Langfuse or LangSmith do that and we don't.

WHAT IT COSTS

One exported request is one event

We count requests, not traces or spans. The free plan runs the same pricing engine as the paid one.

Free

$0

3,000 events/mo · 30 days history · every feature

Pro

$11.99/mo

25,000 events/mo · 90 days history · email alerts

QUESTIONS

Before you paste anything

Do I need to install the SDK?

No. The gateway sends us what we need. The SDK is for apps that call providers directly, and for LLM Gateway.

Does this slow down my requests?

No. Your gateway exports the trace after the response has returned. We're never in the request path.

Do you store my prompts?

No. LLMtrack reads token counts, the model, your labels and timing. Any prompt or completion text in a trace is dropped before anything is written, and it never appears in logs.

Whose cost do you show, mine or the gateway's?

Your gateway's. We record what your gateway charged for each request, and keep each provider's published price next to it, so you can see whether you're paying more. If your gateway doesn't report a cost, we use the provider's published price.

What if I forget to label a request?

It still counts and is still priced. It shows under "unlabelled" until you add a feature, so nothing goes missing from the total.

I also send some of these calls through the LLMtrack SDK. Will they count twice?

No. Pass the provider's response ID to track() as responseId, and LLMtrack matches the two copies, keeps one row with the details from both, and counts it once. Calls that only look alike, with no ID to compare, are labelled "possible duplicate" and kept out of your totals until you decide. Hidden copies still count toward your monthly events, so sending responseId is the setup that keeps everything exact. You choose how both cases behave in Keys & Integrations.

My gateway isn't listed.

If it can export OpenTelemetry traces to a URL, point it at https://otlp.llm-track.com/v1/traces with your key in the Authorization header. Tell us which gateway at support@llm-track.com and we'll write its guide.

Keep your gateway. See where the money goes.

One URL in your gateway settings, and your next request shows up on your dashboard already priced.

Connect your gateway free →