Integrating LLM engines in your Delphi app with Chatterbox

But the real changes happened under the hood.

I’ve refactored the codebase to fully separate the AI client logic from both the UI and the network layer. And what does that mean for you? It means you can now use the Chatterbox AI libraries directly in your own code.

Let me say that again, loud and clear:

You can now use Chatterbox AI libraries from your code!

To test this, I created a very basic demo called EngineDemo. And naturally, for this announcement, I had to explain how the demo uses the Chatterbox libraries.

Since this was a brand-new development that the AI had no prior knowledge of, I decided to revisit an experiment I mentioned in a previous post—one that was only partially successful. This time, I took a different approach. But more on that in the next post.

For now, take a look at what ChatGPT had to say on the topic. Just keep in mind: this wasn’t a simple “click and publish” process. ChatGPT and I had a long back-and-forth, and the result of that collaboration is the article below.

Delphi and AI [7]: How good are local DeepSeek models (for Delphi)

Due to potential privacy concerns with DeepSeek servers (we’re unsure if the data sent over the paid API is kept private), I looked into some smaller DeepSeek models available on the Ollama.com site. These models use less complex AI with fewer parameters than the online version, but they might still be good enough for an average Delphi programmer. We’ll see.

For testing, I used a powerful RTX 4090 card with 24 GB of memory. If your graphics card has less memory, your selection of useful models will be more limited.

I asked all models the same two questions: one on a general programming topic and another specific to the FireMonkey platform. The first question was:

"I have a multiline string containing newline ASCII characters (TMemo.Text). I want to change it to a single-line string with only printable ASCII characters. I could do that with BASE64 encoding, for example. I would, however, like to keep the text as much readable as possible by "encoding" only non-printable characters. Is there a simple way to do that?"

You can check Codellama’s response in an older post (Codellama being the only local model I had tested so far): Delphi and AI [5]: Encoding Multi-line Strings.

Last week, I asked the same question to the online DeepSeek-Reasoning model. Check the answers in this post: Delphi and AI [6]: DeepSeek-Reasoning Model.

The second question was:

"How can I copy text to clipboard in a Delphi Firemonkey application?"

Read here for Codellama’s response: Delphi and AI [4]: Device-ndependent clipboard.

The answer from the online DeepSeek-Reasoning model can be found at the end of this post.

Let’s see how the models are performing! As always, full logs are available on GitHub.

Chatterbox v0.2 - Now with DeepSeek support!

As I’m writing this, a new Chatterbox release is available on GitHub. As with the initial release, you can either download the source and recompile it or grab an executable (Windows, 32-bit) from the release.

(If you have never heard of Chatterbox, you should probably read the initial release article.)

The biggest change since v0.1 is support for DeepSeek AI.

Delphi and AI[6]: DeepSeek reasoning model (Encoding a multi-line string)

An interesting AI has just appeared on my radar - DeepSeek. It exposes a "normal" chat model deepseek-chat and a "reasoning" (that's the interesting part) model deepseek-reasoning. While working on DeepSeek support for the Chatterbox (more on that in few days) I thought it would be interesting to ask the "reasoning" model the "multi-line string encoding" question that every other AI has failed.

Warning: DeepSeek privacy policy states: "When you use our Services, we may collect your text or audio input, prompt, uploaded files, feedback, chat history, or other content that you provide to our model and Services." Don't send any private or proprietary information in the chat!

The log is provided on GitHub.

Just as a reminder, the question was:

"I have a multiline string containing newline ASCII characters (TMemo.Text). I want to change it to a single-line string with only printable ASCII characters. I could do that with BASE64 encoding, for example. I would, however, like to keep the text as much readable as possible by "encoding" only non-printable characters. Is there a simple way to do that?"

Delphi and AI[5]: Encoding a multi-line string

Recently I had to convert a multi-line string into a single-line string value for storage (ignore the 'why' of it; let's just blame it on a legacy code). My first idea was to do a Base64 encode of the string, but I was in a mood for some fun and so I asked my friendly AI helpers:

"I have a multiline string containing newline ASCII characters (TMemo.Text). I want to change it to a single-line string with only printable ASCII characters. I could do that with BASE64 encoding, for example. I would, however, like to keep the text as much readable as possible by "encoding" only non-printable characters. Is there a simple way to do that?"

Let's see what they came up with!

Full transcripts, as usual, are on GitHub. This time I had also created a program containing all implementations.

Delphi and AI [4]: Device-independent clipboard

While working on Chatterbox I ran into number of problems, most of them caused by my limited knowledge of device-independent programming with Firemonkey. One particular challenge was copying data to the clipboard - something that is AFAIK available on all supported platforms. So I asked my friendly AI helpers:

"How can I copy text to clipboard in a Delphi Firemonkey application?"

As always, all logs are available on GitHub.

Delphi and AI [3]: The meaning of Christmas

It's the time of the year to be merry and have fun, so no code today!

I've decided to ask our future AI overlords about the meaning of Christmas. The Delphi way, of course!

Today I'm posting just the answer I liked the most. You can read the others on GitHub.

Happy holidays, fellow programmers! And remember - Delphi, it is the way!

As befits an article on AI-generated content, the image was created by Midjourney.

Delphi and AI [Intermission]: Introducing Chatterbox

t’s great that Embarcadero has added an AI Chat window to RAD Studio, but let’s be honest—the implementation is lacking. As one participant at a recent workshop noted: “It looks like someone wrote it two hours before the release.” Sadly, I have to agree. The AI Chat feature is practically useless if you intend to use AI for anything more than a quick demo.

That’s when I started considering better ways to interface with LLMs. Instead of searching for an alternative, I decided to build my own AI chat interface. It’s called Chatterbox, and it’s fully open source. You can find it on GitHub under a “do with it (almost) whatever you want” license.

To start using Chatterbox, you can either build it from source or download the precompiled EXE from GitHub (currently available only for Windows 32-bit). The app is written with FireMonkey, so—at least theoretically—it can also be built for macOS, iOS, Android, and Linux. However, I haven’t tested it on platforms other than Windows.

To build Chatterbox, you’ll need the following libraries: Spring4D, DCPCrypt2, and TAES. Links to these dependencies are included in the README file.

Delphi and AI [2]: Clipboard Monitor

While preparing for my Delphi and AI workshop, I decided to keep a log of all my interactions with AI helpers in a file for later analysis. Initially, I searched for an existing utility to log clipboard changes to a file (and I found one), but then I thought—why not ask the AI helpers to help me create one? After all, it’s not a big problem: set up a timer, check if the clipboard content changes, and log the content to a file. What could be simpler?

I posed the same question to all five engines:

"I want to create a Delphi application that would monitor clipboard content (on Windows) and append clipboard content to a log file each time the clipboard has changed (and has a text inside)."

Let's see how they performed!

Logs and code are available here.

Delphi and AI

Recently, I led a workshop in Slovenia where we explored the current state of AI in relation to Delphi programming. (A note to participants: the slides are finally online—apologies for the delay!) The initial results were, let’s say, interesting enough to warrant further study.

Now, let’s see how today’s "state-of-the-art" AIs perform with Delphi programming!

The contenders are:

• OpenAI with the o1-mini model (my testing suggests that it gives better results than chatgpt-4o)
• Ollama with codellama:13b model (the largest model I can run on my NVidia 4090)
• Gemini with gemini-pro (currently pointing to gemini-1.5-pro)
• Claude with claude-3-5-sonnet-latest (at this moment this resolves to claude-3-5-sonnet-20241022)

As I care for the privacy of the code I send to these tools, I'm using a paid version of Gemini.

These four engines I'll be using directly from the RAD Studio, most of the time through the "AI Chat" panel. 

When the privacy wouldn't be a concern, I'll also be using:

• CoPilot via Microsoft Edge (I use it a lot and I like it for general-purpose questions with googlable answers)

All chat logs will be published on GitHub.