Disclaimer:
This blog post reflects my thoughts on Windsurf at the time it was written. Since then, Windsurf has undergone significant changes, ...
For further actions, you may consider blocking this person and/or reporting abuse
How cascade is different then copilot edits?
It has full access to your code base and can create new files, edit, correct, and learn from your files without having to pass a huge context into it's prompt.
All you need to do is describe what you want done, and it will look over the files, fix them, and add anything that is necessary.
Plus it's free!
In my opinion, the implementation is still quite weak, even with a paid PRO period, comparable to Cursor.
I share my experience using both IDEs based on the Claude 3.5 Sonnet model
Cons:
When you directly specify a folder for the context, it selects only partial files and does not see everything else, even when you specifically ask it to analyze all the files in the folder, so the context is very broken.
Autocomplete is very poorly implemented compared to Cursor, and also offers very stupid autocompletes, for example, deleting working pieces of code that the question had nothing to do with.
It often hallucinates, supplementing the code with non-existent pieces of code that have nothing to do with logic at all and creates incorrect imports or non-existent ones at all.
There is no or I did not find a setting specifically for Codeium AI, where you can set a prompt instruction, for example, answer in a certain language, etc. This is strange, because when installing Codeium AI as a separate extension in VS Code, these settings are present.
There is no option to select beta features, such as greater context, like in Cursor in 200k, using Claude. Ready for much more than $10+ for such additional features.
Pros:
Has very good potential due to the use of searching for specific files through imports for better context, but this needs improvement for now.
No bugs, like in Cursor, with loss of context when pasting parts of code directly into the chat.
Summary:
So far, a very raw product that needs significant improvements, but has good potential due to specific search by project.
Overall, I would like to say that I like it, but there are still many minuses. I would like to have beta features such as greater context or analysis of all files in a specified directory, not just those that it chooses itself.
I will follow the product from time to time and wish the developers success.
Thanks for your detailed comment! I appreciate you taking the time to compare both products. While I agree with most of your points, there are a few areas where I have a different perspective:
I haven't encountered this issue myself. I frequently use this feature without problems. The only time I faced a similar challenge was when I asked it to recursively convert all files within a directory (over 20 files spread across more than 4 folders) from one framework to another. It processed some files but missed others. My workaround was to convert each folder individually, which worked perfectly. I wonder—would Cursor handle this scenario better?
I agree. The autocomplete could definitely be improved.
In my experience, hallucinations have been rare, though they do happen occasionally. This might vary depending on the programming language. Personally, I treat the AI like a junior developer—I explain tasks clearly and correct any mistakes it makes. I find that issues arise when people expect the AI to read their minds or handle complex tasks without clear instructions.
I agree; this would be a great addition. Hopefully, they implement it soon.
As you mentioned, the app is still in beta, so there's plenty of room for improvement. Ultimately, I hope people see AI tools as collaborative aids rather than replacements for themselves.
Thanks for the reply.
Yeah, maybe I should describe prompts more descriptive, cuz the issue with not seeing some implemented methods in references files makes me sad.
But in general they reeeealy have a good potential and this the first tool after cursor I'm using as Pro-paid
I still can't figure out what full, 1000 steps/month means for Cascade. Is that a 1000 chat limit per month?
Nope. Even with the free tier, you have unlimited chats.
The steps I believe are more of the level or depths.
I'm not really sure, 😂, but I can say for sure it has nothing to do with the chatting
Couldn’t get it to work with .net maui app. Seems the Microsoft .net Maui extension was filtered out in list of extensions . The meteor extension is full of bugs so doesn’t work . Please add the default Microsoft .net maui back
I don't work for or with the makes of this product.
I was working on pos system using AI in cursor. The free prompts were all used. So, i downloaded WindSurf, because I heard it’s completely free but this too has limit for generating. The limit has reached and I was unable to complete the project. The cursor is far better than WindSurf but prices are high for the people like me, who just want to create projects just for personal use.
not free anymore
It's still completely free if you use their base model, which is what I am currently using until I pay for Pro again.
Unfortunately as of today, only chatting us free, write mode is no longer free no matter the model. Hence the end of this free AI tool for devs.
Is the pro-tier worth it?
But I'd suggest starting off with the free tier first to see if you like the experience.
Definitely. I know after trying tons of AI tools this is the only AI editor that is both fairly priced and actually works properly.
You can try for yourself.
wheres link to dl this?
Hey sorry, forgot to add it to the post. Here you go
codeium.com/windsurf
gist.github.com/jugrajsingh/a31f9d...
I created a quick install script
wheres link to dl this plz?
codeium.com/windsurf