Technology never stops, and the ecosystem is evolving too fast! Today we’re going to take advantage of the fact that Embarcadero has released a new version of KAI , version 1.1.1 , which brings improvements to the use of local models. In this post, we’ll show you how you can combine these two worlds: the total privacy and zero cost of local AI with the power of KAI integrated directly into your RAD Studio .
The Video: Complete Practical Tutorial
For those who like to “learn by doing” and want to visually follow each click and setting, I’ve prepared a detailed video showing all of this in action. It’s a complete step-by-step guide to ensure you don’t make any mistakes configuring Ollama with KAI.
Watch the video below for details:
In the video, you’ll see everything from the Ollama service running on the machine to the actual interaction within RAD Studio , where the model analyzes a complete project and returns precise information. It’s fantastic to see the IDE working independently with the local model!
Setting up the environment: Ollama + RAD Studio
Setting up your local environment is quick and easy. I’ll show you how to prepare your IDE right now:
Step 1: Ollama Verification First of all, make sure Ollama is active. You can search for it in the Windows Menu or check the icon in the system tray. To validate, open your browser and access:
[crayon-6abc19cb27043810662498/]
If a message appears confirming that Ollama is running, we’re ready for the next step!
Step 2: IDE Configuration With RAD Studio open, go to the menu: Tools > Options > KAI > Chat .
Step 3: Connection and Template On the Caixete tab , we will configure access:
- In the address field, paste the URL:
http://localhost:11434. - Click the connection test button.
- Once connected, KAI will list the installed models (in my example, I used Gemma 412B ). Select your model and click save. (Tip: If you use LM Studio , the logic is the same, just point to the address and port it provides!)
What has changed in KAI 1.1.1? (Technical Updates)
Version 1.1.1 is a weight maintenance update, focused on stability and expanding the use of local models. Check out what’s new:
- Enhanced Integration: Improved support for the Ollama and LM Studio engines , ensuring much smoother communication between the IDE and the AI.
- Powerful “Agentic” Flows: KAI now supports longer sequences of tool calls . This means that AI can perform complex tasks, such as reading multiple files, analyzing context, and intelligently suggesting changes in sequence.
- Greater Reliability: Significant improvements in creating new files and modifying existing units within your project.
- MCP Interface Refinements: The MCP Tools configuration page now displays long descriptions correctly, and the “Select All / Deselect All” behavior has been refined.
- Visual Adjustment: The display of column headers in MCP tools when using the Dark Theme in the IDE has been corrected. Now everything is legible and looks great!
[crayon-6abc19cb2704e041234963/]
Practical Demonstration: “Describe the Project”
To test the power of this version, I ran a sensational test analyzing a real project called “certificado client,” a FireMonkey app for Android . I opened the KAI chat , selected the local session with the Gemma 412B template , and sent the command:
“Describe the open project. I need to know what it’s about.”
The result was impressive! Because it’s local processing, the hardware is pushed to its limits—you’ll hear the machine’s cooler working hard! In my test, it took about 4 minutes to process. That might seem like a lot compared to the cloud, but remember: it was 100% offline , with complete code privacy and zero token cost! KAI perfectly identified it as a Delphi app for 64-bit Android and detailed its functionalities. Awesome, right?
Advantages of the On-Premise vs. Cloud Model
Still in doubt? Check out this quick comparison:
| Feature | Local Model (Ollama/LM Studio) | Cloud Model (OpenAI/Anthropic) |
|---|---|---|
| Cost | Zero cost (uses your hardware) | Paid by token (monthly or prepaid) |
| Privacy | Total. The code doesn’t leave your machine. | Data processed on external servers. |
| Speed | It depends on your local CPU/GPU. | Usually faster |
| Connection | Works 100% offline | Requires a stable internet connection. |
| Settings | Requires configuring the local engine. | It only requires the API key. |
How to Update and Conclusion
To run this version now, open your RAD Studio and go to GetIt Package Manager . Search for KAI and install version 1.1.1 . Remember: the official release was on September 15th, but general public availability began on September 25th .
This update is a great step for those who want the best of AI without compromising the security of their data.
I hope you enjoyed this new feature. If you liked it, you know what to do: give it a thumbs up, share it with your development team, and subscribe so you don’t miss anything from the Delphi world!
Big hug, best of luck with your coding, and see you in the next post!
[crayon-6abc19cb27052087448250/]