Honest things I found after 30 days testing AI companion apps

Hi, I'm Souvik, I run an independent AI companion review site.

Three things I genuinely did not expect to find:

Memory is mostly fake

Most platforms market "persistent memory" but context resets when you close the browser. I planted 10 specific details in week one and checked what each platform recalled at day 30. Most got 3 or 4 right. One got all 10. That platform is also one of the cheapest. The expensive ones were not the ones with better memory.

"Uncensored" usually isn't

There are two ways to build a no-filter platform. Remove restrictions at the model level, or intercept responses at the application layer before showing them to you. Most platforms claiming to be uncensored are doing the second thing. You find out weeks in when you hit a wall mid-conversation with no warning.

The real monthly price is not the advertised price

Several platforms have a $9-15 headline subscription but layer credits on top for images, voice and video. I calculated real costs for a moderate user. On some platforms the actual monthly spend was three times the advertised price by month two.

That's what I've been documenting at companaya.com happy to answer questions about anything in the category.