PyRomusa AI — v0.11.0 🤖

• Version type: STABLE
• Release date: 2026-10-24

Overview

This version/framework focuses on creating a very simple, beginner-friendly chatbot system, built with:

• custom logic (Chatbot() object)
• real machine learning models (RealChatbot() object)

It exists to:

• help beginners understand how chatbot logic can work
• allow fast experimentation with simple AI-like behaviour
• avoid complex frameworks and heavy dependencies (Chatbot() object)

What makes it different:

• you can avoid neural networks, hidden layers, ML libraries (Chatbot() object)
• you can using LLM tehnologies (RealChatbot() object)
• fully custom, readable logic

This version implements two types of chatbot objects:

• Chatbot() (optimized for few training examples and hardware-undemanding)
• RealChatbot() (optimized for large, serious projects, real experimentation and hardware-demanding)

Files Included

General structure (exactly this structure will be found if you search in 📁all versions of this repository):

|-📁PyRomusa_AI - vX.Y.Z/
|----📄README.md
|
|----📁PyRomusa_AI/
|-------- 🐍PyRomusa_AI.py
|-------- 🐍errors.py
|
|-------- 📁Datasets/
|------------ more Python files...
|
|-------- 📁Reply_Engines/
|------------ more Python files...

More exact structure (if you search in the 📁src folder):

|- 📁src/
|--- ⚙️pyproject.toml
|--- 📄README.md
├─── 📁pyromusa_ai/
    |--- 🐍core.py
    |--- 🐍__init__.py
    |--- 🐍errors.py
    ├─── 📁Datasets
    └─── 📁Reply_Engines

Where:

What's New

• Docstrings for all available functions: Now, to find out in an instant what a specific function does and what role each of its parameters plays, you simply need to hover your cursor over that function; you will then see a description of the function along with all the available parameters. (works on VSCode)

• |RealChatbot() object| Raw text as training example: Do you want your chatbot to learn how the language you’re training it on works, or perhaps you simply lack question-answer pairs? Well, you can now use the bot.trainer.start() function: set the example_type parameter to "raw" and provide your raw text in the training_input_example parameter—it’s as simple as that.

• |RealChatbot() object| Timer on training loop: Find out how many seconds your training session lasted by setting the show_total_time_of_training_when_ends parameter to True. The result will appear when the training session ends.

• |RealChatbot() object| Censored words system: This system has actually been in place within Chatbot() for some time now. If you aren't yet familiar with it, its purpose is to replace certain words with others, thereby helping to reduce the use of vulgar language in the responses.

• |RealChatbot() object| New parameters added (weight_decay & optimizer_type) in bot.trainer.start() function: For greater flexibility, you can now change the optimizer type (currently AdamW or Adam) for training. The weight_decay parameter serves the exact same purpose as it does in PyTorch optimizers.

• |RealChatbot() object| Specified value type for all parameters available on bot.trainer.start() function: Each parameter here should accept only one type of value (integer, string, float, or boolean). To prevent future errors and avoid running code unnecessarily, you can now receive an error before the code executes, allowing you to correct the mistake immediately.

• |RealChatbot() object| New tokenizer type option, "semiwords-level": Are you looking for something between a word and a character—a balance between the quality of the responses and the flexibility of generation? Well, this tokenizer groups about 2–3 letters (on average) into a single token.

• |RealChatbot() object| A optional feedback system (with y/n) while you are talking to the chatbot: This system is recommended if you call the bot.reply_at() function multiple times during a single execution. It serves to refine the chatbot's final responses based on user preferences by calculating the similarity between the most recently generated response and a list of previously rejected responses. If the similarity exceeds the value defined by the max_allowed_similarity_procent_with_banned_replies parameter (a float representing a percentage), the system will attempt to regenerate the response until one with a lower similarity score is produced, subject to a specified maximum number of attempts.

• |RealChatbot() object| Clear separation between special tokens and normal tokens: In case you didn't know, special tokens form an important part of the training text structure, while normal tokens are those entered by the user to help the chatbot converse more naturally. Well, chatbots can now generate special tokens directly and fully; this positively influences the quality of the generated response, making it less likely that the chatbot will say things unrelated to the prompt (though this also depends on the training examples).

• |RealChatbot() object| Removing unnecessary comments from the code: To keep the source code cleaner, I have removed many unnecessary comments. This does not, in any way, affect the practical use of the framework.

• Update to the Romanian language dataset, 'Teacher for PyRomusa AI': It has been added more training examples to be able to answer more prompts. It is now a better dataset than in the previous version.

✅ STABLE release notice:

• The API is considered stable and ready for regular use
• Core logic is implemented and tested
• Behaviour is consistent across typical use cases
• Minor bugs may still exist, but no breaking changes are expected

Quick Usage Examples

1. Creating your first functional chatbot ever

from pyromusa_ai import Chatbot

# Create a chatbot
bot = Chatbot(chatbot_name="RomusaBot")

# Add training data
bot.trainer.add_data("Hello!", "Hi there!")
bot.trainer.add_data("Bye!", "Goodbye!")

# Start training
bot.trainer.start()

# Get a response
print(bot.reply_at("Hello!"))

2. Learn to start using the framework

from pyromusa_ai import Chatbot

# Create a chatbot
bot = Chatbot()

# Get help
bot.helper.how_to_start()

Note:

Depending on how you installed PyRomusa AI, this framework, in your code, must be imported like this, one of these 2 variants:

a) If you installed with pip install ...:

from pyromusa_ai import Chatbot

b) If you installed manually from this repository, preserving file names, from 📁all versions folder:

from PyRomusa_AI import Chatbot # If your code is located in the same folder as 🐍PyRomusa_AI.py

or

from PyRomusa_AI.PyRomusa_AI import Chatbot # If your code is in the same folder as the folder that has all the resources for PyRomusa AI (📁PyRomusa_AI)