We built CaraML because we were tired of switching between AI platforms. But once everything was in one place, a new problem appeared: which model should I actually use for this?
That question is what led us to build Flavia.
The Problem with Choice
Every AI model is good at something different. But the landscape shifts constantly — a model that was the best at coding last month might have been overtaken by something new this week. Keeping track of which model is best for what is practically a full-time job.
Most people just pick one model and stick with it. That works, but you're leaving a lot on the table. The right model for a creative brief is not the same as the right model for debugging code or summarizing a research paper.
How Flavia Works
Flavia sits between you and the models. You type your prompt, and Flavia figures out which model is the best fit — considering the type of task, the complexity, and what's likely to give you the best result. It all happens instantly, before the prompt is even sent.
It also learns from how you work. Over time, Flavia picks up on your preferences and adjusts. If you tend to prefer longer, more detailed responses for certain tasks, Flavia factors that in.
Staying Fast, Staying Private
Model selection has to be instant — if it adds noticeable delay, it's not worth it. Flavia's routing is lightweight enough that you won't feel it. It just works.
And because this all happens on your side, your prompts aren't stored or used for training. Flavia gets smarter for you without your data going anywhere it shouldn't.
Why We Built It
We didn't want CaraML to just be a hub where you manually pick between models. That still puts the burden on you. Flavia is our answer to that — it removes the decision fatigue so you can focus on what you're actually trying to do.
The best tool is the one that makes the right call without you having to think about it.