A short explanation, followed by a path to the full guide. Start with the term you ran into; you don’t need to learn them all at once.
Agent
A system that uses a model to choose some actions and next steps toward a goal.
API
A defined interface through which software requests data or an action from other software.
Automation
A process that runs steps when a trigger or schedule occurs. It may use no AI at all.
Benchmark
A shared test for comparison. Its score may not predict performance on your actual task.
Connector
An integration that lets an application access another service within granted permissions.
Context window
The bounded amount of information available to a model in a request.
Embedding
A numerical representation used to compare content or other inputs.
Evaluation / eval
A test of whether a system meets defined quality and behavior criteria.
Fine-tuning
Additional training that changes a model’s learned parameters using selected examples.
Foundation model
A broadly trained model that can be adapted or used for different tasks.
Generative AI
Systems that produce content such as text, images, audio, or code.
Governance
The people, rules, evidence, and decisions that make AI use accountable.
Guardrail
A control intended to constrain behavior. It needs testing and can fail.
Hallucination
An output that sounds plausible but contains unsupported or incorrect information.
Human in the loop
A person who can review or intervene at a meaningful point in a system.
Inference
Using a trained model to produce an output from new input.
LLM
A large language model: a model trained to work with language, often alongside other modalities.
MCP
Model Context Protocol: a standard for connecting AI applications to tools and contextual data.
Memory
Information an application preserves and makes available later. It is a designed feature.
Multimodal
Working with more than one type of information, such as text, images, and audio.
Orchestration
Coordinating steps, models, tools, or agents, including state and failure handling.
Plugin
A packaged extension that adds capabilities to an application; its contents vary by platform.
Prompt injection
Untrusted input that tries to redirect an AI system’s behavior or authority.
RAG
Retrieval augmented generation: finding relevant material and using it as context for generation.
Skill
Reusable task instructions, sometimes bundled with examples or code, in a supporting application.
Structured output
A response constrained to a format or schema. Valid structure does not establish truth.
Token
A unit used to represent input and output for a model. Tokenization depends on the model.
Tool calling
A model requesting an available function; the surrounding application executes and controls it.
Vector search
Finding nearby numerical representations, often to retrieve semantically related material.
Workflow
An organized process with inputs, steps, outputs, and handling for exceptions.