I Got Tired of Surprise OpenAI Bills, So I Built a Dashboard to Track Them

The author received an OpenAI bill about three times larger than expected and lacked visibility into the cause.

They built LLMeter, an hourly polling system that normalizes usage from multiple providers into a Postgres database.

LLMeter encrypts API keys, retries failed API calls with exponential backoff, and breaks down costs by provider and model.

After a month, LLMeter found one background job using gpt-4o caused about 70% of costs, saving roughly $200/month after a fix.