Share My Creation B4XCopilot - A.I. Assistant

Blueforcer

Well-Known Member
Licensed User
Longtime User
Magma said:
That means... you can update your tool with cli.. copy code and paste to a jshell prompt cli... also with the new feature coming from Erel ..codesync will change everything!!
Click to expand...
oh nice didnt saw the codesync feature announcement..
 

Magma

Expert
Licensed User
Longtime User
@Blueforcer
can you share what model worked best for you (at free)...
and some tips working with cli
 

Blueforcer

Well-Known Member
Licensed User
Longtime User
Magma said:
@Blueforcer
can you share what model worked best for you (at free)...
and some tips working with cli
Click to expand...
Gemini 3 (flash) is my all day LLM. At heavy work or problems im switching Gemini 3 pro. Just use it like a coworker . Im currently writing a gemini cli skill wich handles the work in B4X a bit better. For further discussions we should open a new thread in "chit-chat"
 

Blueforcer

Well-Known Member
Licensed User
Longtime User
Magma said:
coworker - what is - and how ?
Click to expand...
I mean, just like your colleague at work, just tell him what to do and he'll figure it out. The more information you give him at the start (e.g., look in XYZ.bas and fix this and that), the better.
 

Magma

Expert
Licensed User
Longtime User
- You mean be more specific - what 2 do...

- found something - using settings.json (At windows) from users/%userfolder%/.gemini/settings.json you can set some more settings (like mcp server, or how to compile fast to check if something works)
B4X: 
{
  "mcpServers": {
    "slackCoworker": {
      "command": "python",
      "args": ["slack_mcp.py"]
    }
  }
}
 

Magma

Expert
Licensed User
Longtime User
at trusted folders - you have only the project folder you want to search ?
 

cheveguerra

Member
Licensed User
Longtime User
If you are using Gemini or ChatGPT via CLI to analyze your project, I highly recommend creating a file named 00_instructions.txt in your source folder.

Most CLI tools read files in alphabetical order. By naming it 00_, you ensure the AI reads your "Rules of Engagement" before it processes any code. This "Sequential Priming" significantly reduces hallucinations and prevents the AI from suggesting Java-style code that won't work in B4A.

B4X: 
''' ======================================================================================
''' CRITICAL INSTRUCTIONS FOR THE AI ASSISTANT (B4A PROTOCOL)
''' 1. CONTEXT: This project is developed in B4A (Basic4android).
''' 2. ANTI-HALLUCINATION: Before suggesting properties, methods, or syntax, you MUST
'''    VERIFY if they exist in the current project's [Global Variables] or [Types] definitions.
''' 3. EXTERNAL DOCUMENTATION:
'''    - IF files with names containing "Library" or "Reference" are provided, PRIORITIZE those definitions over your general knowledge.
'''    - Use these files as the "Source of Truth" for B4A classes and methods.
'''    - IF NO references are provided, rely strictly on the existing code within the project modules.
''' 4. SCOPE & SYNTAX:
'''    - Strictly respect Public/Private visibility.
'''    - B4X is case-insensitive; maintain consistency but do not treat case as a difference.
'''    - Avoid suggesting Java or VB.NET specific syntax unless it is native to B4X.
''' ======================================================================================

For the part of External Documentation:

To make this even more powerful, I also include a second reference file containing the definitions of the internal B4A libraries (Core, SQL, OkHttp, etc.).

Since IAs are heavily trained on native Android (Java), they often try to suggest native Java methods. By providing a "Library Context" file, the AI is forced to follow the instruction: "PRIORITIZE those definitions over your general knowledge."

This ensures that:
  • The AI only suggests methods that actually exist in B4A.
  • It respects Read-Only [Prop:R] properties, preventing it from suggesting code that won't compile.
  • It follows the exact B4X syntax for events and objects instead of generic Java/Android alternatives.
Make sure your reference files have 'Library' or 'Reference' in their filename so the AI can easily link them to Rule #3 of the instructions.

It’s the difference between an AI that just "reads your code" and an AI that actually "knows how to program in B4A".

Since my english is not that good, this message was mostly written by Gemini, so if you see a lot of formality and "I am too good to be true", then it is not my fault!!
 

Attachments

  • _B4X_Librerias_Internas_para_ayuda_IA.zip
    93.2 KB · Views: 79
Last edited:

cheveguerra

Member
Licensed User
Longtime User
Mashiane said:
Thank you so much, this is very helpful. I appreciate. Also can you please share this resource if you dont mind?
Click to expand...
Jajajajaj sorry, I thought that I had already attached it!!!

I just edited the previous post!!
 
Last edited:

Blueforcer

Well-Known Member
Licensed User
Longtime User
Gemini handles B4X extremely well. I've never had real problems with it, and it never uses native Java unless B4A doesn't support a function natively and then it uses perfect inline java. Most of the system prompt is standard procedure, which Gemini already knows.

I work on my 85k lines of B4A code every day with Gemini. I also tried GPT Codex 5.3 and Claude Opus 4.6 for several days and burned millions of tokens in testing. Gemini is currently one of the best LLMs for B4X in my opinion.
For example, I "vibe coded" a complete B4A HTTP Server from scratch with upload and websocket support, with zero intervention, it runs at the first start perfectly.

Anyway, your system prompt is a good starting point. BUT:

The internal library context will burn your tokens like crazy (calculated 584.088 Tokens wich results in $5,58 with the first promt using opus 4.6 and $2,36 with gemini 3 pro as an example )
where most of the informations are not needed. You should spin up an agent or skill for an intelligent search function to minimize input tokens. Also it's better to use markdown files instead of txt. There are lot of characters wich are unnecessary.
 
Last edited:

cheveguerra

Member
Licensed User
Longtime User

So, the part of the prompt is "probably" helpful, but the external library not so good??

When you say "burned millions of tokens" ... how much (in $$) was that??
 

Blueforcer

Well-Known Member
Licensed User
Longtime User
I really don't recommend using the external library context. If needed, give Gemini a little help with a short example.

For my testing, I started with a standard $20 subscription each, which I pushed pretty fast to the usage limit because I spun up multiple agents in a defined coding orchestra, so I used the API for further processing. All in all, I spent around $160 in 5 days
 

Magma

Expert
Licensed User
Longtime User
well start using gemini cli.. seems works better than claude...

how can i be more specific what will fix,optimize for me... some times saying to it fix one sub and changing a lot...

also how can i go a step back.. is it have history? .. also if go back returns tokens?
 

Blueforcer

Well-Known Member
Licensed User
Longtime User
Just write your promt more specific Promt engineering is everything in vibe coding.
You dont have a history, you should review the changes gemini wants to make before accepting.
You can tell gemini to just undo the last changes, of course it uses also tokens.
 

Mashiane

Expert
Licensed User
Longtime User
Greetings...

I have been swapping between Antigravity (renews tokens on average of 30 minutes to 2 days on my subscription when i finish them), OpenCode : Kimi 2.5 Free (come tomorrow after token expiry) and now CodeX (this has not expired as yet). All have their strengths and weaknesses, At least im able to access my working folders and chat with the tools to do 1,2,3. Managed to get Gemini CLI installed yesterday and will check it out today. Thanks for the heads up.



 

cheveguerra

Member
Licensed User
Longtime User
If you feel the CLI is getting too 'creative' and changing things it shouldn't, just ask Gemini or ChatGPT to write the prompt for you.

You can tell it something like this:

'I need to modify [this specific Sub] in B4X to do [X task]. Write a prompt for Gemini CLI that is extremely specific, strictly forbids it from touching any other part of the code, and ensures it only outputs the modified block.'
Click to expand...
It's much easier to let the AI talk to the AI. This way, you get the surgical fix you need without the 'collateral damage' in the rest of your module.

I know it sounds a bit silly, but it's much easier to ask the AI for a prompt first. Once you see how it builds them, you'll start realizing exactly 'what' and 'how' to ask for things yourself.

You can also ask the IA to explain WHY that prompt works!!
 
Last edited:
You must log in or register to reply here.

Similar Threads

  • Article
Android Tutorial B4X Forum Search Tool - Command Line Search for AI Assistants & Developers
Replies
0
Views
648
Jack Cole
  • Question
Android Question edge to edge problem - assistance will be highly appreciated
Replies
0
Views
638
Zeev Goldstein
  • Question
Android Question need assistance with firebase report
Replies
1
Views
1K
Erel
  • Question
Android Question Test B4XCopilot
Replies
2
Views
2K
Blueforcer
  • Question
Android Question [SOLVED] Need assistance with RSA decryption
2
Replies
26
Views
19K
Johan Schoeman
Menu
Log in
Register
Cookies are required to use this site. You must accept them to continue using the site. Learn more…