Skip to content

3.7 — Vocabulary Bank (Phase 3)

TermDefinition
AgentAI system with tools, decision-making ability, and multi-step execution capability
Tool use / Function callingMechanism letting AI take actions beyond generating text — read files, call APIs, run commands
MCP (Model Context Protocol)Standard protocol for connecting AI agents to external tools and data sources
MCP ServerProgram that exposes tools from an external system to an AI agent
HookScript that runs automatically when a specific event occurs in the harness
RAG (Retrieval-Augmented Generation)Retrieving relevant information from a knowledge base and injecting it into AI context
EmbeddingNumeric representation of text used to measure similarity between documents
Vector databaseDatabase optimized for storing and searching embeddings
Semantic searchSearch by meaning rather than exact keyword match
GuardrailA constraint placed on an agent to limit its actions — a safety boundary
Human-in-the-loopDesign pattern where an agent pauses for human approval at critical decision points
AutonomyThe degree to which an agent can act without human direction
OrchestrationCoordinating multiple agents or systems to accomplish a goal
PipelineA sequence of processing steps where the output of one becomes the input of the next
InferenceWhen an AI model processes input and generates output. Every response = one inference call. Inference costs money.
LatencyTime between sending a request and receiving a response — the delay
ThroughputHow many requests a system can handle in a given time period
Token budgetThe maximum tokens allocated for a task or session — controls cost
DeterministicSame input always produces the same output. Traditional code is deterministic. AI is not.
Non-deterministicSame input can produce different outputs each time. AI models are non-deterministic.

Previous vocabulary: Phase 2 — Vocabulary | Next vocabulary: Phase 4 — Vocabulary

Phase overview: Phase 3 | Next phase: Phase 4 — Orchestrator