I ran out of credit as indicated. I bought flex credits and they lasted less than a day, much less than the 500 in the plan. Now I don't want to buy flex again because they last very little and I can't renew the plan until the renewal date. I also don't want to go up to the 60 USD plan, which is a lot considering the slow progress. The program does a lot of strange things that you may have evidenced:

When the code doesn't work and asks for explanatory code, it seems to be confused with a lot of text or code input.

Sometimes he makes drastic decisions without being asked to do so by the user. For example, to solve errors, he deleted part of the code without asking and simplified it, something totally contrary to what was requested, delaying the process and even more so, charging credits for the rework.

It iterates frequently, trying the same solutions that haven't worked before. There isn't some kind of internal log to tell you that's not the way to go.

If you know how to improve this I will thank you. Every time I ask to save the project I have to be extremely specific, but when I close the App, for example, I have to explain the entire project in great detail again, which causes me to lose credits.

Please, the Windsurf team, solve these issues, I liked this much more than Cursor, but this makes it difficult and costs the planned progress.

Instead of fixing the errors directly from debug console. create test suites. cascade performs error fixing far more robustly, saving flow credits and not disrupting the integration of your codebase.

The errors should be handled like Lovable.dev does because i lost lots of credits fixing errors only.

Support told me that this: it is not happening, no way you are getting charged for errors.

BS, it charges me, and I am out of here!

Absolutely agree

I agree. I also watched this part carefully, and the AI ​​does not recognize more than 200 lines. And it does not look at these 200 lines carefully. It just looks at a few lines here and there and works, and if there is a problem, it goes back and looks at a little more and works like that, so errors keep occurring. This needs to be improved. Even if the user explains it carefully, the requirements are not properly accepted.

If the code is more than 400 lines it fails 90% of the time and you have to make multiple request and ask it to update it in chunks and all this counts towards the 500 requests.

It would be better if you can add cursor type option of adding your own API key so we can use the features we need and doesnt impact the cost for codeium.

I like the product but keep getting frustrated due to this limitation and even if i upgrade to 60$ plan it wont solve this issue. Please fix this bug on priority. Windsurf should be able to read and process long codes without limitations, this will help you scale compared to your competitors

codium should credit people able to submit accurate bug reports that lead to actionable fixes. They should absolutly have a bug bounty program, real dollars for security issues and credits for user experience issues.

This is another reason to allow for a local LLM integration, you could interact and generate code with a local LLM then potentially use Claude or Open AI to review your code base and make suggestions to optimize etc... I would still be ok with paying for the $10 a month for the updates and the basic 500 prompts etc.. but Ollama integration would make your prompt budget go farther!