Terminology

• Model: The underlying AI system (e.g., gpt‑4o, gemini‑1.5, claude‑3).
• Context Window: Maximum tokens the model can consider at once (prompt + output).
• Token: Sub‑word units used for counting context and billing.
• Temperature: Controls randomness; higher is more creative but less deterministic.
• Top‑p / Top‑k: Sampling methods to control diversity of outputs.
• System Prompt: Instruction that sets assistant behavior/persona.
• Tool Calling: Let the model call functions/tools with structured arguments.
• MCP: Model Context Protocol for connecting external tools in a standard way.
• Provider: A service exposing AI models (OpenAI, Anthropic, Google, etc.).
• API Key: Secret credential to authenticate with a provider.
• Embeddings: Vector representations of text for search/retrieval.
• TTS: Text‑to‑Speech to read responses aloud.