# The AI terms that matter. Source: https://customlabs.io/glossary/ Updated: 2026-09-13 Glossary # The AI terms that matter. Plain-English definitions for the vocabulary behind our work: retrieval, agents, evals, cost. Each one links back to the insight, capability or tool that goes deeper. ## Foundations ### Retrieval-Augmented Generation (RAG) Also known as RAG Retrieval-Augmented Generation (RAG) retrieves relevant passages at query time and feeds them into an LLM's context. [Read the full entry →](https://customlabs.io/glossary/retrieval-augmented-generation/) ### Embeddings Also known as vector embeddings Embeddings are numeric vectors that place similar content close together in vector space. [Read the full entry →](https://customlabs.io/glossary/embeddings/) ### Context Window The context window is the maximum text, measured in tokens, a model can consider at once. [Read the full entry →](https://customlabs.io/glossary/context-window/) ### Token A token is the basic unit of text an LLM reads and writes. [Read the full entry →](https://customlabs.io/glossary/token/) ### Model-Agnostic Architecture A model-agnostic architecture lets an application swap or route between LLM providers without a rewrite. [Read the full entry →](https://customlabs.io/glossary/model-agnostic-architecture/) ### Fine-Tuning (vs RAG) Fine-tuning trains a model's weights on your own examples, changing its behavior directly. [Read the full entry →](https://customlabs.io/glossary/fine-tuning/) ## Retrieval ### Vector Search (Semantic Search) Also known as semantic search, similarity search Vector search finds the nearest matches to a query by comparing embeddings, not exact keywords. [Read the full entry →](https://customlabs.io/glossary/vector-search/) ### Chunking Chunking splits source documents into smaller passages before embedding, so retrieval returns focused text. [Read the full entry →](https://customlabs.io/glossary/chunking/) ## Agents ### Agent (Agentic AI) An agent is an LLM given a loop, memory, and tools it can call. [Read the full entry →](https://customlabs.io/glossary/agent/) ### Tool Calling Also known as function calling Tool calling lets a model request a structured action instead of only generating text. [Read the full entry →](https://customlabs.io/glossary/tool-calling/) ### Model Context Protocol (MCP) Also known as MCP MCP is an open standard for connecting LLM applications to tools and data sources. [Read the full entry →](https://customlabs.io/glossary/model-context-protocol/) ### Context Engineering Context engineering decides what occupies a model's context window at every step, and in what order. [Read the full entry →](https://customlabs.io/glossary/context-engineering/) ### Tool Contract A tool contract is the schema a tool exposes to a model. [Read the full entry →](https://customlabs.io/glossary/tool-contract/) ### Idempotency An idempotent operation produces the same result no matter how many times it runs. [Read the full entry →](https://customlabs.io/glossary/idempotency/) ### Structured Output Structured output constrains a model's response to a defined schema instead of free-form prose. [Read the full entry →](https://customlabs.io/glossary/structured-output/) ### Guardrails Guardrails are the checks that keep an LLM or agent inside acceptable bounds in production. [Read the full entry →](https://customlabs.io/glossary/guardrails/) ### Golden Path A golden path is the single sanctioned way to do a common piece of work. [Read the full entry →](https://customlabs.io/glossary/golden-path/) ### MCP Gateway An MCP gateway is a shared layer in front of several MCP servers, centralizing controls. [Read the full entry →](https://customlabs.io/glossary/mcp-gateway/) ### Context Rot Context rot is a model's declining ability to use context well as it fills with stale material. [Read the full entry →](https://customlabs.io/glossary/context-rot/) ### Compaction Compaction summarizes or drops older context to make room in a filling window. [Read the full entry →](https://customlabs.io/glossary/compaction/) ### Context Isolation Context isolation gives each delegated unit of work its own scoped context window. [Read the full entry →](https://customlabs.io/glossary/context-isolation/) ## Evaluation ### Eval Suite (Evals) Also known as evals, evaluation suite An eval suite is a versioned set of test cases that measures whether outputs are good. [Read the full entry →](https://customlabs.io/glossary/eval-suite/) ### LLM-as-Judge LLM-as-judge uses another LLM to score a model's outputs against a rubric, at scale. [Read the full entry →](https://customlabs.io/glossary/llm-as-judge/) ### Observability Observability captures traces of every prompt, retrieval, tool call, and response for debugging. [Read the full entry →](https://customlabs.io/glossary/observability/) ### Hallucination A hallucination is a confident, fluent output that is factually wrong or unsupported. [Read the full entry →](https://customlabs.io/glossary/hallucination/) ### Prompt Injection Prompt injection is untrusted input crafted to override a model's system prompt or task. [Read the full entry →](https://customlabs.io/glossary/prompt-injection/) ### Shadow Deployment Also known as shadow testing, dark launch A shadow deployment runs a candidate release against real production traffic, offline, before release. [Read the full entry →](https://customlabs.io/glossary/shadow-deployment/) ## Cost & ops ### Inference Cost Inference cost is what it costs to run a trained model on a request. [Read the full entry →](https://customlabs.io/glossary/inference-cost/) ### Notebook-to-Production Notebook-to-production is the gap between a working prototype and a system that runs reliably. [Read the full entry →](https://customlabs.io/glossary/notebook-to-production/) ### Canary Release Also known as canary deployment, canary rollout A canary release exposes a change to a small, deliberately chosen slice of traffic first. [Read the full entry →](https://customlabs.io/glossary/canary-release/) ### Output Drift Also known as model drift, quality drift Output drift is a gradual change in a live system's output, with no release behind it. [Read the full entry →](https://customlabs.io/glossary/output-drift/) ### Model Deprecation Model deprecation is a provider retiring a model version on its own schedule. [Read the full entry →](https://customlabs.io/glossary/model-deprecation/) ### Acceptance Rate Acceptance rate is the share of an agent's proposed changes a human reviewer approves. [Read the full entry →](https://customlabs.io/glossary/acceptance-rate/) ## Security & governance ### Data Processing Agreement (DPA) Also known as DPA A DPA is the contract naming a vendor as a processor of personal data. [Read the full entry →](https://customlabs.io/glossary/data-processing-agreement/) ### Data Residency Data residency is where data is physically processed and stored, not where users are located. [Read the full entry →](https://customlabs.io/glossary/data-residency/) ### Red Teaming Red teaming is deliberately attacking your own AI system to find what breaks first. [Read the full entry →](https://customlabs.io/glossary/red-teaming/) ### AI Register An AI register is a current list of every system that touches a model. [Read the full entry →](https://customlabs.io/glossary/ai-register/) ### High-Risk AI System A high-risk AI system is one classified as carrying enough decision impact to trigger heavier compliance. [Read the full entry →](https://customlabs.io/glossary/high-risk-ai-system/) ### Human Oversight Human oversight is a checkpoint where a person can meaningfully approve, reject, or intervene. [Read the full entry →](https://customlabs.io/glossary/human-oversight/) ### System Card A system card is a short document stating an AI system's purpose, limits, and owner. [Read the full entry →](https://customlabs.io/glossary/system-card/) ### Shadow AI Also known as shadow use Shadow AI is a model or tool used without going through the sanctioned path. [Read the full entry →](https://customlabs.io/glossary/shadow-ai/) ### Tool Poisoning Tool poisoning hides an instruction in a tool's description that the model reads as legitimate. [Read the full entry →](https://customlabs.io/glossary/tool-poisoning/) ### Confused Deputy A confused deputy is a component tricked into using more authority than a request deserves. [Read the full entry →](https://customlabs.io/glossary/confused-deputy/)