Most devs think they have AI costs under control because they set a $500 hard cap on OpenAI.

tbh, that's not cost control. That's just setting a timer for when your app goes down.

Here's the problem: when you hit $400 and get that warning email, what do you do? You look at the dashboard and see a massive spike in GPT-4o usage. But who caused it?
Was it your new enterprise client onboarding? (Good, increase the limit).
Was it a free-tier user who figured out how to loop your agent? (Bad, ban them).
Was it a bug in your own RAG pipeline retrying the same chunk? (Very bad, fix the code).

The provider dashboard won't tell you. It just shows a giant wall of tokens.

If you are building a multi-tenant SaaS, you need to map every single LLM call to a specific user ID. If you can't attribute the cost, you don't know your unit economics. You might be losing $2 on every $1 you make from your power users.

We built LLMeter exactly for this. It's an open-source dashboard that sits between your app and the provider. You pass the user ID in the metadata, and we track the cost per user, per day, across OpenAI, Anthropic, DeepSeek, and OpenRouter.

No proxies, no Stripe lock-in. Just raw data.

You can self-host it (AGPL-3.0) or check it out here: https://llmeter.org?utm_source=devto&utm_medium=article&utm_campaign=2026-04-25-devto-cost-to-user-mapping