Agents glossary
Learn the meanings of agent-related terms.
Definitions of agents-related terms. Refer to the main glossary for definitions of terms across Retool.
A
Action
An operation within an agent that accepts an input and returns a result. Some examples of actions include a single JavaScript function, or a single call to an LLM.
Agent
A system that can take actions based on large-language model (LLM) reasoning. Agents can call tools (for example, workflows and queries) to gather information and complete or delegate actions.
Agent-to-agent (A2A)
Agent-to-agent (A2A) protocol is an open standard developed by Google that provides secure communication between AI agents, allowing for collaboration and exchange of information in order to complete complex tasks.
Agentic workflow
A workflow where the control flow is influenced by the output of an LLM.
B
No glossary entries found.
C
Configuration Assistant
A feature of agents that uses an LLM to automatically write instructions and create tools for your agent.
Core tool
A prebuilt tool that Retool provides with an agent.
Custom tool
A type of tool for an agent that leverages reusable, complex logic with strongly-typed parameters. They are constructed using a canvas of pre-built query blocks that you connect together to interact with resources, transform data, and run additional logic.
D
Dataset
A collection of records that are passed into an agent when running an eval. A dataset contains name, description, number of items, created, and updated properties.
E
F
Function generator
A chat interface that uses an LLM to automatically create a custom function for use by an agent.
G
No glossary entries found.
H
Human-In-The-Loop (HITL)
A process that requires human action in order to proceed, such as a tool call that requires approval, or Assist resource query creation.
I
Invoke Agent block
A type of block that triggers an agent from a workflow.
J
No glossary entries found.
K
No glossary entries found.
L
M
Module Context Protocol (MCP)
An open standard introduced by Anthropic that provides a standard protocol for LLM-applications (clients) to call tools from a remote system (server).
N
No glossary entries found.
O
No glossary entries found.
P
No glossary entries found.
Q
No glossary entries found.
R
Reviewer
A mechanism that scores the output of an agent eval. Add reviewers when creating a dataset item for an agent eval.
S
No glossary entries found.
T
Tool
Tools allow agents to take actions in other systems. A tool can be a workflow, a query, an MCP server, or another agent. An agent can call a tool to fetch data, perform computations, or interact with other parts of the system.
U
No glossary entries found.
V
W
No glossary entries found.
X
No glossary entries found.
Y
No glossary entries found.
Z
Zero-based numbering
Wikipedia: Zero-based numbering is a way of numbering in which the initial element of a sequence is assigned the index 0, rather than the index 1 as is typical in everyday non-mathematical or non-programming circumstances. Under zero-based numbering, the initial element is sometimes termed the zeroth element, rather than the first element; zeroth is a coined word for the ordinal number zero. In some cases, an object or value that does not (originally) belong to a given sequence, but which could be naturally placed before its initial element, may be termed the zeroth element. There is no wide agreement regarding the correctness of using zero as an ordinal, as it creates ambiguity for all subsequent elements of the sequence when lacking context.