Hi Team,
Is it possible to plug in our own AI model’s API key inside the Vibe Coding block, so that our AI can act as the assistant, connect with the database, and communicate with it directly? I know the Ask AI feature exists, but that runs on a credit basis, whereas here we’d like to build a Vibe Coding app using our own API key so that we fully own and control everything.
Does this make sense?
This would be great indeed.
@akshatbandlish @Tuur this will come soon along with generic rest API integration too
Thanks Artur!
I do want to share some additional (hopefully constructive) feedback about the vibe coding / AI block experience in general.
I think this is the way forward. For straightforward tasks, it works well and can be very powerful. But in practice, I am noticing that a lot of AI credits are being consumed by things that ideally should already be documented.
A significant part of the credit usage (in my case at least) seems to go to:
asking what is actually possible in the AI block environment
going back and forth to understand limitations
searching for workarounds when the AI suggests something that turns out not to be possible
correcting repeated attempts the AI already made earlier in the same chat
cases where the AI gets stuck in a loop and the user has to stop the process without a useful result → credits down the drain?
These are familiar issues when working with AI in development, but here the main concern is that users are spending paid credits inside a kind of black box just to understand the boundaries of the system.
That is the part that feels off to me.
I can see the credit history, but not in enough detail to really understand where credits are going or which interactions were actually productive. I went through the credits of our Business plan surprisingly quickly, without getting the same level of practical output in return.
So while I think the feature is promising, I believe there is room for improvement.
What would help a lot:
deeper documentation for advanced users about standard functionality, constraints, and limitations
clearer guidance on what the AI can and cannot do in this environment
more detailed insight into credit usage
better safeguards against repetitive loops or stuck reasoning
My overall conclusion is positive: I believe this is the future, and I’m glad Softr is building in this direction.
But users should not have to spend credits to discover undocumented limitations. Better documentation and better transparency would make the experience much stronger and much fairer. And to be honest, an AI documentation pipeline could drop release notes with included dev details as soon as a new feature goes live, right?
Less Vibe, More Guide?