VS Code + Claude experience

D

Didier99

Member
Licensed User
I started using Claude Code with VS Code a week ago. It has been a life changing experience. The project was migrating an embedded system from an 8051 to an ESP32, while (of course) adding new features. I did in a weekend what I had anticipated would take 1-2 months (of working at home on weekends and evenings) and it was pleasant, unlike rewriting code you have already written. The refactoring alone was absolutely on point. I would guess a 10x to 20x multiplier compared to writing code the old fashion way (you know, the way you actually do write code yourself like we used to do in the old days). I would expect the multiplier to increase once I get better used to it.

Claude code is easy to read, documented, functionally and logically broken down and organised, and typically runs the first time. If something is missing, I tell Claude what's missing and it has very rarely taken more than one fix for each step.

And I am barely scratching the surface (no skills or agents yet.)

For those who have not tried it yet, you ought to.

Interesting anecdote: I wanted to archive the conversation I had with Claude because it was so fascinating. So I googled it. Google came back with a relatively onerous procedure where I had to download and configure another extension, and check several files by myself to see how they were setup. I started that and quickly gave up. Instead, I asked Claude to archive the conversation in an html file with the same color coding as the conversation itself in the terminal. Which Claude did. I checked the html file and it was what I expected. When it was done, I told it that I may want to do that again so to streamline the process. Claude created a python program that automatically archive the current conversation for any project I am working on. Voila!
 

MichalK73

Well-Known Member
Licensed User
Longtime User
How did you find CloudeCode's pricing for your project?
I considered using it, but the cost of writing a small Python app was about $15.

I've now purchased DeepSeek 4 (DS4) for $5 as a trial. I generated an API key and connected it via a plugin designed for DS4 that connects to VSCode's Copilot. I had them write the same app as before. It was a bit different, but it worked. It cost 25 cents.

Yesterday, I tried this method to see if it worked with B4X in VSCode and the plugin: https://www.b4x.com/android/forum/threads/b4x-vs-code-b4x-ide-companion-%E2%80%94-a-complete-tutorial-for-cross-platform.170777/
Surprisingly, DS4 handles our B4X code remarkably well. In my large project with ABMaterial, I asked him to add new features: buttons, edit fields, and implement the logic according to the specified rules. He did everything well. He placed the code where it should be and created almost perfect user logic. Perhaps if I had been more specific, the results would have been better. Overall, it was a positive experience.
 
D

Didier99

Member
Licensed User
I use the $20/month tier. I have been using it pretty heavy for the last 2 weeks so I went into the app to see how I was using my tokens and such, and the 3 bar graphs were at 0%!!! I would be just fine with a $0.20/month plan!
For now, I would rather not use the Chinese tools, like Deepseek or Qwen, even though they are relatively well regarded.
I am looking at using Gemma4, the free Google tool you can download locally.

Over the weekend, I used Claude to help me with the migration of my main website, where the OS and all the standard utilities were well past the freshness date. It went really well, even though my workflow was definitely not as smooth as VS Code (details below). There was a lot of obsolete php code that had to be updated. The only real issue was migrating the ~200GB of data between the two hosts. The old host and the new one could not agree on a commonly supported security protocol for rsync so we ended up using a less secure method, which was not a problem for me since all the data on my website is freely downloadable anyways But it did take 20 minutes before we could get that started. Once running, it went at about 25MB/s which was pretty good. Another example where I had been pushing this farther because I knew it would have been several weekends of work. It was done in about 4 hours, and 3 of those were the file transfer itself.

While waiting for the file transfer, we worked on a web app I started working on 2-3 years ago but never finished. Well, that's pretty much done now too
At the rate this is going, I might actually be able to retire at some point!

Workflow on the website: I use WinSCP over ssh to edit the files locally through my favorite text editor and WinSCP does the file transfer both ways in the background. This is actually pretty nice, but Claude has no integration for that, even though I am thinking of asking Claude to write one for me so that he could do the work instead of me... So I used the Claude Chatbox to ask questions and ask it to fix bugs or implement new features, it would tell me what commands to run on the server and paste the result (through a Putty box), occasionally it would create entirely new modules that I would upload via WinSCP. Lots of cut and pasting was required, but it got the job done. I even got compliments from Claude about my code
 
Last edited:

Mashiane

Expert
Licensed User
Longtime User
Didier99 said:
I am looking at using Gemma4, the free Google tool you can download locally
Click to expand...
You can do that by installing ollama and install the claude cli too.

B4X: 
ollama launch claude --model gemma4

This will start your claude coding agent and use gemma. To use the cloud version of that, just replace gemma with "gemma4:31b-cloud", this is how I am working on these. Other options are using LLMStudio but that does not fit in vscode.

Its also important to check the LLMArena leaderboard to find what could potentially work best for you based on user votes.




PS: In the B4X IDE Companion VSCode Extension, launching these is already built in. you just give it a model name after you select the action to perform.



Remember: You do have an option to turn off settings for your code base not to be used for training other models. Apparently with Ollama this is default [I need to investigate that though to be 110% sure about such a claim], so I stand to be corrected.
 
D

Didier99

Member
Licensed User
I wish I could like this more than once Thank you!

While we are at it, at work we can only use OpenAI through Azure (the corporate version that does not have the IP leaks of the cloud versions). Do you know if it
is possible to similarly use the Claude agent with OpenAI in VS Code? (recognising that the Claude agent may be blocked for me anyway because of the well-known US government spat with Anthropic)
 
You must log in or register to reply here.
Menu
Log in
Register
Cookies are required to use this site. You must accept them to continue using the site. Learn more…