LLM
Large language modelA model trained on large amounts of data to process and generate language.
AI Notes / Field guide
24 AI terms.
A clearer picture.
From language models to connected agents.
Models & inputs
SLM
Small language modelA language model with fewer parameters, often requiring less compute. “Small” has no fixed cutoff.
Prompt
The input that guides a model’s response: instructions, questions, examples or other context.
Tokens
Units a model processes. For text, a token may be a word, part of a word or punctuation.
Context & information
Context window
The token capacity available for a model’s current input and generated output.
Memory
Information an AI system retains and reuses across steps or interactions; it need not change the model itself.
Chunking
Dividing content into smaller pieces for processing, indexing or retrieval.
Embeddings
Numerical vectors that represent features of content, enabling comparisons such as semantic similarity.
Search & retrieval
RAG
Retrieval-augmented generationRetrieving relevant information and supplying it to a model to help ground its response.
Vector database
A database that stores and searches vectors, often embeddings, to find similar items.
Hybrid search
Combining keyword search with vector search to find exact matches and semantically similar content.
Reranking
Scoring retrieved candidates again to put the most relevant results higher in the list.
Tools & connections
Tool calling
A model requests an external capability, such as search or code execution; the surrounding system runs it.
Function calling
A form of tool calling: the model produces a function name and structured arguments for software to execute.
MCP
Model Context ProtocolAn open protocol connecting AI applications to tools and context provided by external systems.
A2A
Agent2Agent ProtocolAn open protocol enabling independent agent systems to exchange information and coordinate tasks.
Agents & workflows
AI agent
A system that uses a model to choose steps and use tools toward a goal, within defined limits.
Agentic workflow
A sequence combining AI steps and tools. Routes may be predefined or include agent-led decisions.
Agentic AI
A broad term for AI systems that pursue goals through actions with some autonomy; usage varies.
Multi-agent system
Multiple agents interacting to perform work, often with different roles or specialisms.
Planning & oversight
Planner
A component or process that breaks a goal into steps and may revise them as work progresses.
Evaluator
A mechanism that assesses outputs or behaviour against criteria, using rules, models or human review.
Guardrails
Checks and controls that constrain inputs, outputs or actions to reduce unwanted behaviour.
Observability
Using logs, traces and metrics to understand a system’s behaviour, performance and failures.