I was considering creating some projects and realised the cost of some AI models was becoming prohibitive. I found there was nothing out there which was focussed on reducing the costs, so I decided to tackle one area: input tokens for natural language AI projects. TokenCrush is basically a maxi data cleaner for natural language. You send it the prompt that you want to send to another AI, it will return an optimized version of this prompt. It gets rid of the fluff and repitition. For some use cases, this can reduce the cost by a large amount. For example, a sentiment analysis app that reads news articles and tries to determine if they're positive or negative for a company. For news articles, TokenCrush will typically reduce the text by 80%, which is a direct saving. TokenCrush itself only charges per token saved, so you know that you will be saving money (assuming your AI is more expansive that $0.2 per million input tokens). Please give it a try. There's a query tool when you log in, to test out the transformations. Each free account is allowed 10k tokens to be saved per day. No need to give credit card info.

TokenCrush

Thrive in a world of increasing LLM costs

Thrive in a world of increasing LLM costs