The agentic AI glossary.
The vocabulary of agent systems, defined in plain language — and what each concept looks like inside a working runtime.
All terms
AI agent
An AI system that pursues a goal by taking actions — calling tools, reading data, making decisions — rather than only generating text.
Agentic workflow
A workflow where one or more steps are performed by a reasoning agent instead of fixed logic.
Autonomous agent
An agent that operates without step-by-step supervision: given an objective, it plans, acts, evaluates its own results and continues until done or blocked.
Multi-agent system
Multiple agents cooperating on work too broad for one — a researcher feeding a writer feeding a reviewer, or specialists coordinated by an orchestrator.
Agent orchestration
Coordinating which agent or tool acts, in what order, with what context — routing tasks, merging results, retrying failures and enforcing gates.
Agent memory
Persistent state an agent carries across sessions: facts about its user, corrections received, patterns that worked.
Agent skills
Packaged, versioned expertise an agent loads on demand — a playbook containing process, standards and examples for one class of task.
MCP (Model Context Protocol)
An open protocol for connecting AI models to tools and data sources through a standard interface.
Tool calling
A model invoking external functions — search, database queries, file writes — by emitting structured calls its runtime executes.
Function calling
The model-side mechanism behind tool use: the model returns a structured invocation — name plus JSON arguments — matching a declared schema, which the host executes and feeds back..
LLM runtime
The environment that turns model output into real work: executing tool calls, managing context and memory, enforcing permissions, retrying failures, recording what happened.
Local-first AI
AI software whose data, credentials and execution live on hardware you control, with cloud as an option rather than a requirement.
Prompt caching
Reusing computation for the repeated prefix of prompts — system instructions, tool schemas, standing context — so consecutive calls pay only for what changed.
RAG (Retrieval-Augmented Generation)
Fetching relevant documents at query time and placing them in the model’s context, so answers ground in your data instead of training memory.
Context window
The maximum text a model attends to in one call — instructions, history, retrieved material and its own output.
Fine-tuning
Continuing a model’s training on domain examples to shift behaviour — terminology, format, tone — into the weights.
Inference
Running a trained model to produce output — the operational, per-request phase of AI, as opposed to training.
System prompt
The standing instruction that defines a model’s role, rules and boundaries before any user input arrives.
Webhook automation
Triggering workflows from HTTP callbacks the moment events happen — a payment, a form, a commit — instead of polling on a schedule.
Human-in-the-loop
Automation that pauses for human judgment at defined points — approving an outbound email, releasing a payment, accepting a diff.
Execution receipt
A structured record an agent produces for each operation: what was read, which tools ran with what arguments, what was decided and what was written.