Layer 2: Pillar - Glossary

Glossary: 30 AI Agent Terms in Plain English

Bookmark this. Updated monthly.

TermSimple Definition
AgentSoftware that perceives, reasons, acts toward goal
ReActReason + Act loop - thought then action
Tool / Function CallingLetting LLM call your API with JSON
RAG (Retrieval Augmented Generation)Search your docs, add to prompt, then generate
Vector DBDatabase that searches by meaning, not keyword
EmbeddingNumber version of text that captures meaning
ChunkingBreaking docs into small pieces for embedding
Memory (short/long)What agent remembers now vs forever
PlanningBreaking goal into sub-tasks
ReflectionAgent critiques own work
CrewAIFramework for multi-agent teams
LangChainLibrary to connect LLMs + tools
AutoGenMicrosoft framework for conversational agents
OpenAI Assistants APIManaged agent API with threads/tools
HallucinationWhen LLM makes up false info confidently
GuardrailRule to stop bad actions
Human-in-the-loopHuman approves before critical action
Prompt InjectionMalicious text that hijacks agent instructions
System PromptHidden instructions defining agent role
TemperatureRandomness knob for LLM
TokensBillable pieces of text (~4 chars)
Context WindowHow much text LLM can see at once
ObservabilityLogs, traces to debug agent
EvalTest to measure if agent succeeded
LatencyTime from input to output
Cost per runAPI + infra cost per task
Fine-tuningTraining model on your data
Zero-shot / Few-shotDoing task without/with examples
Multi-agentSeveral agents collaborating
OrchestrationManaging order of agents/tools

Next: Start Here

Continue Learning: Explore our Start Here Guide, AI Agent Glossary and Free Tools.