From indie devs, for indie devs

We built the tool we
needed when our own
AI bill stopped making sense.

LLMtrack started as a personal fix for a personal problem. Here's why it exists, who it's built for, and what we'll never turn it into.

100%

Bootstrapped

<1s

Core promise

400+

Models priced daily

The origin

A bill that didn't add up

It started the way most of these stories start: with a number that didn't make sense.

A side project had been running for a few weeks — nothing exotic, a couple of AI features bolted onto a small SaaS tool. Then one month the API bill landed and it was several times higher than the month before. Not a gradual climb. A jump.

The natural next step was to open the provider's usage dashboard and find out why. That's where the real problem showed up — not the cost itself, but the complete absence of an answer. The dashboard showed a single number, updated a day late, with no way to see which feature, which model, or which user was responsible.

$0.23 → some number nobody could explain, for reasons no dashboard would show, discovered a full day after it happened.

The existing tools didn't fit either. The well-known observability platforms were built for teams running LangChain pipelines with dedicated ML infrastructure — proxies to configure, trace trees to learn, pricing built around enterprise usage. None of that matched a single developer who just wanted to know: which feature is costing me money, right now, in plain numbers.

So the fix became the product. One async call, sent after an LLM response comes back, tagged with whatever feature name made sense. No proxy in the request path. No new framework to learn. Just a straight answer to a straight question — and it answers in under a second instead of a day.

That's the whole origin story. No big team, no roadmap deck, no enterprise sales motion. One developer's bill stopped making sense, and the fix for that became something other developers could use too.

Infrastructure for people building AI products is different from AI products themselves — once you've wired tracking into your codebase, you don't rip it out for a demo three weeks later.

— the principle LLMtrack is built around

What we believe

Three things we won't compromise on

These aren't marketing lines. They're the constraints every product decision gets checked against.

Real-time, not eventually

A cost dashboard that updates tomorrow is an accounting tool, not an operating tool. If it doesn't tell you what's happening right now, it can't help you stop a spike while it's still small.

Your prompts stay yours

We store token counts, costs, latency, and feature names. We never store prompt content or model output. There's no version of this product where that changes.

Built for one person to set up alone

No proxy to configure. No sales call to book. No framework lock-in. If a solo developer can't get from signup to seeing real data in under 2 minutes, we consider that a bug.

Who this is built for

Built for builders, not buying committees

We'd rather be exactly right for a smaller group than vaguely useful to everyone.

✓Solo developers and small teams shipping AI features into a real product, who need to know what those features cost without becoming an ML ops engineer to find out.

✓Indie founders watching margin closely — where the LLM bill is one of the largest line items and the gap between "seems fine" and "is actually fine" matters.

✓Anyone who's been burned once by a surprise invoice and decided never to be surprised again.

—Not built for teams that need deep prompt evaluation, dataset management, or LangChain-native tracing — Langfuse and LangSmith do that well, and we'd rather say so than pretend otherwise.

How we check the numbers

Where every number on this site comes from

Model rates come from each provider's published schedule. Our price sheet re-syncs them every day, and the same sheet prices every event in your dashboard. You can read it at /model-prices.

Blog posts quote rates as they were on the day they were checked, and say which day. Each post links to the provider pages it relies on. When a rate or a product changes, we update the post, change its "Last updated" date and say what changed at the top.

Worked examples are arithmetic you can redo: the token counts and the rates are stated, and the result is shown. Screenshots and demo numbers come from a demo workspace filled with seeded test data, and they are labelled that way. They are never presented as a real customer's bill.

Claims about other tools are checked on the vendor's own pages and dated, because they change often. If you find a number that's wrong or out of date, email support@llm-track.com and we'll fix it and note the correction.

What you can count on

Commitments, not promises

Specific, checkable things — not vibes.

A free plan that stays free3,000 tracked events a month, no credit card, no countdown timer. If you're testing whether this is worth your time, that's on us to prove, not something we gate behind a trial.

Pricing that doesn't punish growth$11.99/month gets you the full analytics suite most competitors charge $120/month for. We'd rather have more developers paying a fair price than fewer paying an enterprise one.

No proxy, everYour LLM traffic goes straight to your provider. We never sit in that path. One async call after the fact — that's the whole integration, and it stays the whole integration.