Models

Introduction – Language Models in the Context of AI Agents

Language models form the foundation of an AI agent’s intelligence. They act as the “brain” that processes text, interprets questions, understands context, and generates coherent responses. These models are trained on large amounts of data to learn patterns of human language and apply them to problem-solving, content generation, and automated decision-making.

In Skyone Studio, selecting the right language model is a strategic step. It determines the quality of responses, processing speed, depth of analysis, and even the operational cost. Different models have different capabilities — some are faster and more cost-efficient, while others are more advanced and contextually accurate.

Every agent, Skill, or workflow configuration depends directly on the choice of an appropriate model.

---

Model Categories

Skyone Studio works with two categories of LLMs:

• Embedded (Native): These are models maintained by Skyone and executed in internal environments, using local resources and a private data network. They are recommended for scenarios requiring high control, security, and confidentiality.

Embedded models are available by default upon contracting. They cannot be edited or removed.

• Integrated: These allow the connection of solutions from external providers (such as OpenAI and Anthropic). To use them, you must register your access key (AppKey) in the system. This category is ideal for accessing specialized models without the need to maintain local infrastructure.

OpenAI model configuration example

---

Embedded models available in Skyone Studio

Currently, Skyone Studio provides the following embedded models:

• Gemma 4 e2b: A version that unifies and updates the capabilities of the previous Gemma 3, Granite, and Llama 3.2 models.

• Gemma 4 31B: A high-performance model for tasks requiring greater robustness.

• GPT OSS: A robust model with high textual precision capacity.

---

How to create an Integrated Model

1. Access the side menu and click on “Models”.

2. Click on “Add”.

3. Choose the type: OpenAI or Customized.

4. After selecting the desired option, fill in the fields below to configure your model:

1. Basic Information

• Image: Click on the "Change image" option to upload a new image or use the trash icon to remove it.

• Model name: Type an identification name for the model.

• Description: Enter a brief description of the purpose of this model.

• API Key: Enter your API key.

• Model: Select the specific model you wish to use from the list.

2. Model parameters

Para personalizar o comportamento do modelo, ative a chave Mostrar na seção de Parâmetros Avançados. Você poderá selecionar um parâmetro existente ou criar um novo clicando em + Adicionar.

To customize the model's behavior, toggle the "Show" switch in the Advanced Parameters section. You can select an existing parameter or create a new one by clicking + Add.

Create New Parameter

When choosing to create a parameter, fill in the following information in the modal:

• Name: Define the parameter name.

• Type: Select the data format (Text, Number, or Boolean).

• Set value as required: Check this box if the parameter is indispensable for execution.

• Default value: Enter the value.

• Input type: Choose how the user will interact with this field. If you select "Select," you must configure the choice options:

1. Click the + Add button within the options section.

2. Fill in the Label (visible name) and the Value (technical data).

3. Click Create to confirm the new parameter.

Click on Save model to finish.

---

FAQ – Frequently Asked Questions About Language Models

Which model should I choose for my AI agent?

It depends on your goal: for simple interactions, use faster and more cost-efficient models; for more complex and contextual responses, choose more advanced models.

Can I change my agent’s model after it’s created?

Yes. In Skyone Studio, you can change the model at any time, adjusting cost and performance as needed.

Are larger models always better?

Not necessarily. They offer greater capability but also come with higher costs and longer response times. The ideal approach is to balance performance and efficiency.

Does the model understand multiple languages?

Yes, most advanced models can understand and generate content in multiple languages.

Can the model be trained with my company’s internal data?

Yes, through fine-tuning or by providing context using prompts and knowledge bases.

What are the main differences between the language models?

The choice of a language model depends on the balance between performance, cost, and speed. The primary distinctions between them lie in four fundamental pillars:

1. Context Window

It refers to the maximum amount of data (tokens) that the model can process and "remember" in a single interaction.

• Large windows: Ideal for analyzing extensive documents, full codebases, or maintaining long conversations without losing track.

• Reduced windows: Recommended for specific tasks and direct commands.

2. Sophistication and Accuracy

Reasoning capability varies according to the complexity of the model's training.

• Advanced Models: Possess greater abstraction capacity, generate more natural responses, and better handle cultural nuances and complex instructions.

• Entry-level Models: More prone to hallucinations in advanced logic tasks, but effective for simple classifications.

3. Response Speed (Latency)

The elapsed time between sending the command and receiving the response.

• "Flash" or Small Models: Optimized for low latency, making them ideal for real-time applications such as support chatbots.

• High-Performance Models: Due to the larger volume of processed parameters, they may exhibit a higher response time.

4. Operating Cost

Financial investment is generally proportional to the size and power of the model.

• Cost-effectiveness: Smaller models are significantly cheaper, allowing for a high volume of requests (e.g., $1,000.00).

• High Investment: Cutting-edge models are recommended only when analytical depth justifies the higher cost per token.