AI agents are like well-trained digital team members that help your business by answering questions and solving problems. They work by combining three main parts: a brain that understands questions (language models), a smart filing system that quickly finds relevant information (knowledge processing), and a systematic way of creating accurate answers (RAG pipeline). Imagine having an assistant who has instant access to all your business knowledge and can provide consistent, accurate responses around the clock – that's what AI agents do.
Core Components
Language Model Foundation
Large Language Models (LLMs) serve as the brain of AI agents, enabling them to:Understand natural language queries
Generate accurate, context-aware responses
Adapt to different user tones and styles
Maintain context across the conversation.
Example: When a customer asks about product features, the LLM understands the query's intent and formulates a coherent response.
Knowledge & Data Processing System
Vector Embedding & Processing Transforms information into a format AI can understand:
Converts text into mathematical representations
Enables precise information matching
Captures meaning relationships between different pieces of content
Example: Converting company policies into vector format allows quick, accurate policy lookups.
Vector Database Acts as the AI's organized memory system:
Stores information efficiently for quick retrieval
Maintains connections between related information
Enables fast, accurate information search
Updates dynamically with new information
Example: Storing customer interaction history for quick access during support conversations.
Structured Data Retrieval
Excel and CSV sources can be indexed as retrievable chunks, or loaded into a SQL table through Tables to SQL so the agent can answer questions by querying the table.
Enables agents to generate summaries, compare values, or filter data based on context
Example: A sales manager asks for the top-performing products in the last quarter—the AI Agent queries the connected sales database and delivers the summary.
RAG Pipeline (Retrieval-Augmented Generation)
Retrieval Phase Efficiently finds relevant information:
Rephrases the user's question for better retrieval
Applies knowledge-source scope and permission filters
Searches using semantic or hybrid search
Reranks the results to surface the most relevant chunks
Example: When asked about leave policy, quickly finding specific relevant policy sections.
Augmentation Phase Enriches queries with context:
Combines the user's question with the retrieved, ranked chunks
Places the highest-ranked chunks into the model's context window
Ensures the retrieved context supports a grounded response
Prepares the complete context for response generation
Example: Adding current policy updates to historical policy information for accurate responses.
Generation Phase Creates accurate, helpful responses:
Generates a clear, grounded response
Includes citations that link back to the source content
Applies appropriate formatting
Maintains consistent tone and style
Example: Generating a complete policy explanation with relevant examples and current guidelines.
Tools Tools connect the agent to your systems through connector actions, workflows, MCP server tools, or custom HTTP endpoints. Using these, agents can:
Trigger workflows in both UnifyApps and external systems
Integrate with enterprise apps through out-of-the-box connectors
Support for custom connectors to your home grown applications
Example: You can enable agents to take action on your systems securely using your APIs without the need for MCP servers.
Capabilities Capabilities are built-in functions that you switch on for the agent, such as:
Web Search & Deep Research
Chart & Visualization Generation
Document Understanding & Summarization
Example: An agent can look up recent market trends via web search, generate a bar chart of competitor performance, and email it to your team.
Unify AI Agents combine the power of LLMs, advanced data processing, and action-based integrations to create a flexible, intelligent, and truly helpful digital workforce. Whether answering questions, retrieving structured data, or acting on your behalf, they’re designed to grow with your business.