Was super excited for the Softr AI update but realized I will not be able to use it much… Adding one button to a previously built vibe coded block burned 10% of my monthly credits.
There is no way to select models, understand where usage is going, or optimize. The vibe coding aspect of Softr is pretty much unuseable after 5-10 prompts and trying to build anything serious will plow it to the ground almost immediately 
For reference, one prompt burned all my 50 monthly credits last month.
Similar experience. On a Business plan ($323/month), I burned through the 100-credit limit in about 10–15 prompts.
The vibe-coded block relies on undocumented imports such as:
@/components/ui/xxx@/lib/datasource@/lib/editable-settingsSo this makes ‘just ask ChatGPT/Claude and paste in the results’ a poor option too. I could probably spend credits trying to get the vibe code block to reveal the documentation it uses internally, but I haven’t had the credits to burn…
Yes, I’d come to much the same conclusion. We’re on the Pro plan and therefore get 50 credits, which basically means any live implementation of AI is largely pointless. I now avoid Vibe Coding on any core pages.
Hi everyone
There is already visibility into AI credit usage. In Workspace settings → Plan and Billing → AI credits → Activity History, you can see exactly how credits were spent, including the specific actions and their cost. So there is a way to trace usage and understand what triggered it, even if it’s not always something users check during day-to-day building.
On the usage side, AI actions aren’t tied to simple UI clicks, but to the amount of generation happening under the hood. Because of that, even relatively small changes (like editing a block or adding a button) can sometimes involve more processing than it appears from the outside. That’s usually where the expectation mismatch comes from.
From my experience, using “vibe coding” effectively works best when you treat it as an iterative workflow rather than expecting a single prompt to produce a final result. For example:
I usually start by giving a detailed description of the use case to an external AI tool (like ChatGPT or Claude) and ask it to generate a well-structured, optimized prompt for the vibe coding block.
Then I use that prompt inside Softr to generate the initial version of the block.
After that, I test whether the core functionality works as expected.
If it works, I refine UI/UX with smaller, targeted prompts directly in the builder.
If something is not working correctly, I take the generated code from the “Content” section, send it back to the external AI tool with the issue description, and use it to debug or improve the solution.
I repeat this loop until the block behaves as intended.
In a way, working with AI is a bit different from working with a human, as instead of delegating and expecting interpretation, it’s more about being explicit, structured, and iterative. Small differences in wording can lead to very different outputs, so refining the prompt becomes part of the process itself rather than a one-time instruction. This approach also helps reduce unnecessary credit usage, since each AI interaction is intentional and builds on the previous step rather than starting from scratch each time.
Overall, your feedback is helpful in highlighting how AI usage feels in real workflows, especially during experimentation and iteration phases, so thank you very much.
I’m seeing the same problem. AI credits disappear far too quickly for something that is supposed to support experimentation and iterative building. Even a small change can consume a surprising number of credits, which makes it difficult to test ideas without worrying about the limit. More transparent cost estimates before running a prompt, better model controls, and a larger allowance would make the feature much more practical.
Hi David, thanks for the full reply but I’m not sure that helps much.
It’s like you’ve given us a fancy hammer to use, but you’re suggesting that we still take our work outside and hit it with a brick from time to time, or the hammer will wear out.
What I need is 10x the credits for the same spend, especially if I want to let my users have ‘Ask AI’ or AI generated fields. At 50 credits it’s largely pointless.