- Prompt
- What you type or say to the AI. Clear, specific prompts get better results.
- Model
- The trained AI that does the actual work inside an app. Also called the engine.
- LLM
- Large language model: a model that reads and writes text.
- Multimodal
- Can handle more than one kind of input, like text plus photos or voice.
- Hallucination
- When AI states something false as if it were true.
- Reasoning model
- A model that “thinks” step by step before answering. Slower, better at tricky problems.
- Open model
- A model anyone can download and run on their own computer, like DeepSeek or Qwen.
- Agent
- AI that takes actions toward a goal, not just answers.
- RAG
- Looking things up in your documents first, then answering from them.
- Connector
- A link that lets your assistant use another app, like Runway or Gmail, for you.
- MCP
- Model Context Protocol: the shared plug standard behind connectors. Works across Claude, ChatGPT and more.
- API
- The behind-the-scenes doorway apps use to rent an AI model from its maker.
- Model picker
- A menu inside an app that lets you choose which model answers.
- Training
- How a model learns, from huge amounts of examples, before you ever use it.
Examples are starting points, not rankings or ads. Tools change quickly, so features and prices vary by plan and country.