Glossary of artificial intelligence.

TO

Agent

System that uses a model to decide steps and operate tools under defined rules.

Orchestration

B

Benchmark

Set of tasks, conditions, and metrics used to compare a specific behavior.

Assessment

c

Context cache

Reuse of input prefixes to reduce latency or cost when the provider supports it.

Optimization

c

Context window

Maximum amount of information that the model can consider in a request.

Tokens

c

Quantization

Reduced numerical precision to use less memory and speed up inference.

Local AI

AND

Embedding

Numerical representation used to measure similarity and retrieve information.

Vectors

AND

Assessment

Reproducible test that measures a behavior with examples and a rubric.

Quality

g

Grounding

Connection of the response to retrieved information, tools, or verifiable external sources.

Evidence

H

Hallucination

Generated content that appears plausible but is not supported by the available data or sources.

Reliability

I

Instruction injection

Untrusted content that attempts to divert the system from its rules or legitimate objective.

Security

M

MCP

Protocol for connecting models and agents to tools or data sources through common interfaces.

Integration

M

Multimodal

System capable of processing or generating more than one type of information, such as text, image, audio, or video.

Modalities

R

RAG

Retrieval of external information before generating a response.

Sources

Yes

structured output

Response adjusted to a scheme to be able to validate and process it.

Integration

T

Token

Unit in which the model divides text and accounts for context and usage.

Consumption