A New Era of AI Decompilation: VB Decompiler 12.7 Expands the Capabilities of AI Helper

After extensive research, testing, and technical improvements, I am pleased to introduce the latest release of VB Decompiler version 12.7. This update represents a significant leap forward in code analysis and source code recovery. While the AI Helper was introduced in version 12.6 with the aim to improve decompiled C# code, VB Decompiler 12.7 significantly expands its functionality, making the tool more universal, flexible, and accessible for a wide range of users.

From C# to Visual Basic: AI Helper now supports VB5/VB6

One of the most anticipated and significant new features in VB Decompiler 12.7 is the support for Visual Basic 5.0 and 6.0 within AI Helper. In version 12.6, the AI Helper's functionality was limited to working with C# code obtained during .NET decompilation. Now, thanks to a new implementation, AI Helper can also work with code decompiled from Visual Basic 5.0/6.0 assembler (Native Code) and P-Code executable files, the very ones that are still considered the most difficult to restore in high quality.

AI Helper analyzes the already generated decompiled code (result of VB Decompiler's decompilation) and applies artificial intelligence based on publicly available models for recognizing automatically unrecognized loops (including recovery of For, While...Wend, Do...Loop), logical variable renaming in accordance with their purpose in the code, declaring variables with correct types, simplifying access to data structures, improving readability, and structuredness of the code.

Important to understand: AI Helper does not correct decompilation incompatibilities and issues, especially in cases where the original VB6 code was compiled into native code (assembler) and there are decoding issues. However, even in such cases, AI processing makes those parts of the code that were correctly reconstructed more readable and suitable for analysis.

Performance and stability: AI in a separate thread

Another important improvement, moving AI Helper to a separate execution thread. Previously, in version 12.6, processing code with AI could block the interface, creating an impression of hanging, especially when working with heavy models. In 12.7, this has been addressed. VB Decompiler uses fewer resources. A Cancel button has appeared, allowing you to cancel processing if the AI model gets stuck or takes too long to respond. A crash or hang in an external AI server (Ollama) no longer affects the stability of the VB Decompiler.

This makes working with AI Helper not only more convenient but also safer. You can experiment with different models without risking losing access to main functionality.

Flexibility in selection: support for YandexGPT and other models in Ollama

Version 12.6 recommended using exclusively the qwen3:8b model, which was adapted for all promts and parsing results. This was justified. Qwen3 shows excellent results when recovering code structure and minimally loads the system. In 12.7, the list of recommended models has been expanded: YandexGPT 5 Lite is now officially supported and adapted to work with VB Decompiler. To install it, simply execute the command in Windows console (the Ollama server must be running):

ollama pull yandex/YandexGPT-5-Lite-8B-instruct-GGUF:latest




Moreover, VB Decompiler is now adapted to work with almost any model in Ollama. The parsers have been updated for better interaction with different AI. Specific tags and analysis blocks filtering has been added that may interfere with processing. Now you can experiment with virtually any model if its output appears more useful. You are no longer tied to a single model and can choose the optimal one for your tasks: performance, accuracy, memory usage.

For example, successful tests were conducted with deepseek-r1:8b and gemma3:27b, and our promts and parsers have been adapted accordingly. I still recommend qwen3:8b, but now you can choose the most suitable model for yourself. VB Decompiler does not limit this.

User Interface Improvements

VB Decompiler 12.7 also introduces important improvements in UX. The context menu is now tied to the active tab, rather than the text field focus. This means that Copy, Select All, and other functions will work correctly even if the window temporarily loses focus. The Disassembler tab now always displays addresses. The Add Address setting only affects the decompiler window, this makes analyzing assembly code more familiar and consistent.

In the absence of a local help file or PDF viewer, the program will automatically open online documentation, ensuring continuous access to the manual.

Also in version 12.7, an issue that affected VB6 users has been fixed: when opening forms, the Designer tab did not appear. The problem is now resolved.

VB Decompiler 12.7 is not just an update, but an evolution in the approach to code analysis. Thanks to the expansion of AI Helper for Visual Basic 6.0, support for new AI models, improved stability and UX, the tool has become even more indispensable for reverse-engineering outdated VB6 projects, recovering lost code, analyzing legacy systems, and analyzing malware in antivirus labs.

Although AI cannot perform miracles where there are compatibility issues due to the decompilation process itself, it significantly improves readability and understanding of the code, transforming complex code into structured and maintainable.

Recommendations for getting started

Install Ollama and download qwen3:8b or YandexGPT-5-Lite. Update VB Decompiler to v12.7 to take advantage of the new AI Helper's features. Experiment with models, find the one that best suits your type of task.

VB Decompiler continues to be a leader in its niche, combining the power of local analysis with cutting-edge AI capabilities without losing control over data and processes. If you are new to VB Decompiler, we are always happy to welcome you among our customers.







* Visual Basic, Visual Studio are registered trademarks of Microsoft Corporation.