📚 Glossary Link to heading

🧠 General Terms Link to heading

• LLM (Large Language Model)
  A machine learning model trained on vast amounts of text data to understand and generate human-like language. Examples include ChatGPT, Claude, Gemini, and Mistral.

• Token
  A unit of text (word, part of a word, or symbol) used by LLMs for processing language.

• Prompt
  A textual input given to an LLM to instruct or query it.

• Context Window
  The amount of text (in tokens) an LLM can consider at once.

• Fine-tuning
  Customizing a pre-trained LLM on specific data to specialize it for a task or domain.

• Embedding
  A numerical representation of text used for semantic understanding, clustering, and retrieval tasks.

• RAG (Retrieval-Augmented Generation)
  A technique where external data is fetched and combined with a prompt to help the LLM generate more accurate responses.

• Chain-of-Thought Reasoning
  A prompting technique that encourages the model to “think step by step” before answering.

• Hallucination
  When an LLM generates factually incorrect or fabricated information with confidence.

• Agent
  A software entity that uses an LLM and tools to perform a task autonomously or semi-autonomously.

• Memory
  Persistent storage of past interactions or user-specific data, allowing an agent or LLM to build continuity.

• Tool Use
  The ability of an LLM or agent to use external APIs or tools (e.g., calculator, browser) to complete a task.

• Multi-agent System
  A setup where multiple AI agents collaborate, often using LLMs, each with different roles or goals.

🚀 Mainstream LLMs (as of 2025) Link to heading

• GPT-4 / GPT-4 Turbo – OpenAI
  GPT-4 Turbo is faster and cheaper than GPT-4 with a 128k token context window. Powers ChatGPT.

• Claude 3 – Anthropic
  Known for long context windows and safety features.

• Gemini 1.5 – Google DeepMind
  Formerly Bard. Integrates with Google apps. Very large context support.

• Mistral – Mistral AI
  Open-weight models such as Mistral 7B and Mixtral.

• LLaMA – Meta
  Series of open-weight models, including LLaMA 2 and upcoming LLaMA 3.

• Command R – Cohere
  RAG-optimized models, popular in enterprise use.

• Grok – xAI (Elon Musk)
  Integrated with X (formerly Twitter). Focused on reasoning and open-source.

🛠️ CrewAI-Specific Terms Link to heading

CrewAI is a Python framework for building collaborative LLM agents.

• Crew
  A team of agents that work together to accomplish a task.

• Agent (in CrewAI)
  An LLM-based worker with a defined role, goal, tools, and behavior.

• Role
  Describes the function of the agent, e.g., “Researcher” or “Writer.”

• Goal
  The specific objective assigned to the agent.

• Backstory
  Optional narrative that gives personality or behavioral context to an agent.

• Tools
  External functions or utilities agents can use (e.g., search, code execution).

• Process
  The workflow strategy — sequential (agents act one after another) or collaborative.

• Task
  A work item assigned to an agent with instructions and output expectations.

• Output File
  Where the agent or crew stores results for reuse or review.

• Memory
  Stores past outputs or steps to inform current and future agent reasoning.

🤖 AutoGPT / AgentGPT Providers and Frameworks Link to heading

Frameworks that support autonomous or collaborative AI agents using LLMs and tools.

⚙️ Related Ecosystem Tools Link to heading

• LangChain
  Framework for chaining prompts, memory, and tools together. Widely used in agent systems.

• LlamaIndex
  Data indexing layer to connect LLMs with custom or private datasets.

• Toolformer
  A fine-tuning technique to teach LLMs how and when to use tools.

• OpenFunction Calling
  Technique that enables LLMs to invoke external APIs via structured JSON outputs (e.g., OpenAI’s function calling, Anthropic’s tool use).