| 1. The Brain | The Engine | Foundation Model (LLM) | The core model (like Claude or Titan) pre-trained on public data that handles logic and language. |
| The Education | Fine-Tuning | Training the model’s neural brain on specific data to permanently change its style or knowledge. |
| The Personality | System Prompt | The initial “God Mode” instructions that strictly define the Agent’s rules and boundaries. |
| 2. The Memory | The Open Book | RAG (Retrieval-Augmented Generation) | The process of reading fresh, private files (like PDFs) in real-time to answer questions. |
| The Index | Vector Database | A specialized database that stores content by “meaning” (vectors) rather than keywords for fast search. |
| Short-Term Memory | Context Window | The temporary slot where the Agent holds the conversation history and RAG documents while thinking. |
| 3. The Agency | The Traffic Cop | The Agent (Orchestrator) | The logic that decides whether to answer directly, search a database, or execute a tool. |
| The Hands | Tools / Function Calling | Code blocks (APIs, Lambda) that allow the AI to perform actions like “Get Balance” or “Send Email.” |
| 4. The Safety | The Filter | Guardrails | An independent security layer that blocks malicious inputs and hides sensitive data, regardless of the AI. |
Blue-Lab 