| Token |
A model input or output unit; often a word piece rather than a full word. |
| Context window |
The amount of input and output a model can consider in one request. |
| Grounding |
Constraining model answers to provided or retrieved evidence. |
| RAG |
Retrieval-augmented generation: retrieve sources, then generate grounded output. |
| JSON Schema |
A machine-readable contract for JSON structure and field validation. |
| Tool use |
A model selecting and calling an external capability through a defined interface. |
| Prompt injection |
A malicious or accidental instruction that tries to override trusted instructions. |
| Evaluation |
A repeatable method for deciding whether a prompt or model output meets success criteria. |