Skills
"Skills" are reusable functions that enhance agent capabilities in the Studio IA platform. They enable the execution of specific operations—such as calculations, database integrations, or business rules—that the model alone cannot efficiently perform, especially in tasks involving mathematics or structured actions.
A Skill consists of Python code and can be used in automated workflows to provide greater accuracy and control over the user experience.
Types of Skills on the Platform
On the platform, Skills represent capability blocks that define how the AI agent will interact with users and process information. They are categorized into two main types: Embedded (Native) and Custom.
1. Embedded (Native) Skills
These are pre-configured models provided by the platform itself. They are ready to use and cannot be edited or deleted by users. Currently, Skyone Studio offers two embedded models:
RAG Skill: Uses the Retrieval-Augmented Generation (RAG) technique, allowing the agent to fetch external information in real time to enrich its responses to users.
SQL Skill: Allows the agent to interact directly with relational databases via SQL, answering questions and generating results based on structured queries.
These models are ideal for common use cases, ensuring robustness and incorporating best practices by default. Learn more below:
RAG is an advanced artificial intelligence technique that combines information retrieval with text generation. Instead of relying solely on the knowledge acquired during the model’s training phase, RAG enables the AI to consult external sources in real time—such as vector databases or documents—to enrich and contextualize its responses.
Core Components and Competencies in RAG
Understanding the RAG Concept: Knowing how AI models can retrieve external information in real time to supplement their generated responses.
Semantic Search: Applying embeddings and vector representations to identify the most relevant documents for a given query.
Vector Indexing: Using specialized tools such as FAISS, Weaviate, Pinecone, Milvus, among others, to efficiently store and search vectors.
Integration with LLMs: Connecting retrieval mechanisms to language models (LLMs) like GPT, LLaMA, and others, ensuring the final output combines retrieved data with fluent text generation.
RAG Pipeline Construction: Developing the complete workflow—query, retrieval, and response—ensuring that each step is integrated and functions efficiently.
Document Embedding: Converting texts into vectors using models like all-MiniLM, nomic-embed, and others, to populate the vector database.
RAG Frameworks: Utilizing frameworks such as LangChain, LlamaIndex, Haystack, and others to build, organize, and orchestrate RAG pipelines.
RAG Evaluation: Measuring system effectiveness in terms of accuracy, response relevance, and latency (response time), ensuring quality and performance.
Governance and Security: Ensuring RAG workflows comply with standards for confidentiality, data accuracy, and access control.
2. Custom Skills
These are created directly by users. There are two main ways to create them:
From scratch: The user builds the entire logic and custom configuration to meet specific business needs.
By duplicating an embedded Skill: The user copies a native model to use as a starting point, then makes adjustments and customizations as needed.
Custom Skills offer full flexibility, allowing the agent’s behavior to be tailored for unique and more complex use cases.
Create a New Skill
To create a custom Skill:
Go to the side menu and click on “Skills.”
Click “New Skill” or select an existing Skill and click “Customize.”
On the edit screen, fill in the following fields:
Module name: This is the main identifier of the Skill.
Functions: Here you can write one or more Python functions. Each function should clearly receive parameters and return values.
After completing all fields and validating the code, click “Save module.”
Last updated