About
We built the layer that AI billing was missing.
Every team that ships AI features eventually hits the same three problems. First: you're running GPT-4o on requests that a much cheaper model could handle just as well. Second: you have no real visibility into what your AI spend is buying you, just token counts and a total at the end of the month. Third: the invoice arrives and nobody on the team can explain exactly why it's that number.
PromptUnit was built to fix all three. We sit between your code and your AI provider, classify each request by complexity and task type, route it to the cheapest model that meets your quality bar, and give you a clear per-feature breakdown of where every dollar went. One base URL change. No model management overhead on your side.
Our pricing is designed around the same logic: you pay 20% of verified savings. If we don't save you money, you don't pay. The incentives are aligned by design.
The problems we're solving
Model sprawl
As the model landscape expands, picking the right model for each request becomes a full-time job. PromptUnit handles that automatically, against a quality threshold you control.
Pricing opacity
Token counts tell you how much you spent. They don't tell you why, or whether it was worth it. PromptUnit gives you per-feature cost breakdowns and savings estimates in real time.
Surprise invoices
AI costs spike without warning. PromptUnit enforces hourly and daily spend limits, detects anomalies, and automatically downgrades runaway requests before they hit your bill.
What we do with your data
Your requests pass through our servers, which is how routing works. We see the request in transit the same way any proxy does.
Prompt content is never written to disk. We log metadata only: token counts, model used, latency, cost, task type. The actual text of your prompts and completions never touches our database.
Your API keys are encrypted at rest with AES-256-GCM and only decrypted in memory to forward your request. We never use them for anything else.
We do not sell data. Anonymized quality signals (model performance per task type) stay inside PromptUnit to improve routing accuracy for everyone.
Get in touch
Questions about security, integration, or enterprise pricing. We respond to every message.
igal@promptunit.aiSee what you're actually spending on AI.
14-day observation period. No routing changes until you turn it on.
Get started free