Agentic AI arrives for Delphi and C++ Builder

Embarcadero has released Kai, an agentic AI assistant for RAD Studio, an IDE (integrated development environment) for Delphi and C++ Builder. Kai is offered as an extension, which means that by default RAD Studio lacks AI capabilities. The extension provides chat, code completion, and an MCP (model context protocol) server to enable other AI agents to communicate with the IDE. Kai depends on third-partly LLMs (large language models), running in the cloud or locally, and users have to supply their own API key in order to use it. Despite this limitation, Kai is a subscription product costing $249 per developer per year, though free trials are available. Delphi, for which the programming language is also known as object pascal, has a long history. Version 1.0 was released in 1995 for 16-bit Windows, and was at the time the most advanced rapid application development tool available, with a drag and drop visual form designer reminiscent of Microsoft’s Visual Basic (VB), accompanied by a native code compiler which, unlike VB, generated executables that did not require a separate runtime engine. C++ Builder came a little later, with the same visual designer but supporting code in C++ rather than object pascal. Why is Delphi still around after more than 30 years? “It’s fully compiled native code,” presales director Stephen Ball told us. “We’ve got software running core parts of stock exchange systems and high frequency trading, where you need speed and performance and the threading you get through a native application … customers are fed up of hybrid applications that are chewing through so much memory and power consumption.” These are the same advantages Delphi has always had, especially for Windows desktop applications, though today it occupies a small niche, used by just 2.5 percent of developers according to the most recent Stack Overflow survey, enough to be viable but not mainstream. The latest RAD Studio is version 13.1, with new support for native Windows Arm binaries. Users who have not touched Delphi for years will find it still familiar, though since its early days it has acquired a cross-platform framework for Windows, macOS, iOS, Android and Linux, called FireMonkey (FMX), a true 64- bit IDE, high DPI (dots per inch) support, new language features including type inference, inline local variables, ternary operators and more. C++ Builder now uses the Clang 20 compiler. Backward compatibility is generally good though and legacy Delphi code will likely run with little tweaking. Kai itself is a somewhat minimalist response to market pressure for AI capabilities in the IDE. Installing Kai and adding an active subscription enables options for configuring models for agent chat and for code completions, which can be ghost text in the editor, accepted by pressing Tab, or suggestions in a completion list, or in a separate panel. Kai chat is open ended and can generate code, resolve build errors, manage version control, perform file operations, and use other MCP servers. We had a quick look at Kai, with mixed results. The ability to configure Kai for local LLM providers is a strong point, with support for Ollama and LM Studio – though developers will need a powerful PC to get acceptable performance, and plenty of RAM in order to use larger models. We also encountered some rough edges. One query which worked in LM Studio chat failed with an error when tried from Kai chat. We were also amused to find Kai chat (using a local model) offering a Python implementation when we asked for code for a conversion function without specifying the language. We asked Delphi product manager Marco Cantu whether Kai is disadvantaged by the niche status of the language, which gives AI less training data to ingest. He agreed that there have been issues, including AI generating old-style code from Delphi 7 (2002) days and not using new features, but said that things have improved. “Today most of the LLMs can do a good job with generating Delphi code. This tool is also for C++, and on C++ they can do an even better job,” he told us. Are there guardrails around Kai, to protect against prompt injection or other risks? “We assume customers are going to deploy this in the context of an application where humans review code and check what’s being done,” Cantu said. Comments on a webinar introducing Kai are mixed. Some attendees felt that Kai should be included for free with RAD Studio or Delphi subscriptions, since it relies on external LLMs; another argued that “$249/yr is very reasonable compared to the time it’s likely to save me.” Others felt that the Embarcadero team should focus on improving core features of RAD Studio rather than adding AI. Another view is that RAD Studio itself does not need Kai. “In my current development workflow I usually only open the Delphi IDE when I have to fine-tune DFMs [Delphi Form Modules] … 90 percent of my development work happens on the Claude Code and Codex CLIs [command line interfaces],” said a developer who seems well adapted to the idea that IDEs are less necessary when coding via agentic AI. Our brief hands-on with Kai left us thinking it is a work in progress that will likely improve. It was also a reminder that for developing a high performance, easily deployed GUI application Delphi remains an excellent solution.®

المصدر: The Register