AI Agent Glossary: Every Term You Need to Know

The AI agent space moves fast and invents vocabulary even faster. This glossary cuts through the jargon. One concept, one sentence, alphabetical order.

A

Agent -- A program that uses an LLM to complete a task autonomously, deciding its own steps rather than following a fixed script.

Agent chain -- A sequence of agents connected by events, where each agent's completion triggers the next. Also called a pipeline or workflow.

Agent framework -- A library for building individual agents (LangChain, CrewAI, AutoGen). Provides LLM abstraction, tool use, and memory. Not the same as orchestration.

Agent library -- A registry of reusable agent definitions, each with a name, prompt, tools, and configuration.

Agent profile -- A saved configuration for an agent: which model to use, what environment variables to inject, what CLI arguments to pass.

C

Chain -- Short for agent chain. A defined workflow of agents that execute in a specific order based on events.

Chain builder -- A visual or code-based tool for defining chains. Mentiko has both a drag-and-drop canvas and a JSON editor.

Chain runner -- The component that reads a chain definition and orchestrates agent execution. In Mentiko, this is a bash script.

Conditional branching -- When an agent emits different events based on its output, routing the chain down different paths.

D

Decision flow -- Mentiko's AI-assisted decision-making feature. A structured process that helps teams decide between options using a 3-round guided wizard.

E

Emit -- When an agent produces an event upon completion. "The researcher emits research:complete."

Event -- A signal that an agent has completed its work. In Mentiko, events are files written to disk. The event system watches for new files and triggers the next agent.

Event-driven architecture -- A system design where components communicate through events rather than direct calls. Agents don't call each other -- they emit events that the system routes.

Execution environment -- Where agents actually run. Options include local filesystem, SSH remote servers, or Docker containers.

F

Fan-in -- When multiple parallel agents complete and a single downstream agent waits for all of them before proceeding.

Fan-out -- When one agent's completion triggers multiple agents to run in parallel.

I

Iterative loop -- A pattern where an agent produces work, a reviewer agent evaluates it, and the producer revises based on feedback. Repeats up to N times.

M

Marketplace -- A community registry of pre-built chains, agents, and templates that users can browse and install.

Multi-tenancy -- Architecture that isolates data between different teams or customers on the same platform. Each organization sees only its own chains, runs, and agents.

O

Orchestration -- Coordinating multiple agents to work together on complex tasks. Handles sequencing, event routing, error recovery, and monitoring.

P

Pipeline -- Another word for chain. A sequence of agents that process data or complete tasks in order.

PTY session -- A pseudo-terminal session. In Mentiko, each agent runs in its own isolated terminal session, giving it access to CLI tools like git, npm, and python.

Q

Quality gate -- A checkpoint in a chain where output is evaluated before proceeding. If quality is below a threshold, the chain branches to a revision loop or human review.

R

RBAC -- Role-Based Access Control. Defines who can do what: owners, admins, members, and guests have different permissions.

Run -- A single execution of a chain. Each run has a status (running, completed, failed), timing data, and per-agent logs.

Runbook -- A predefined set of steps for handling a specific situation. In agent chains, runbooks are encoded as agent prompts.

S

Scheduling -- Running chains automatically on a cron schedule, via webhook trigger, or on-demand.

Secrets vault -- Encrypted storage for API keys, database credentials, and other sensitive values. Injected into agent sessions at runtime.

T

Trigger -- The event that causes an agent to start. "The summarizer triggers on research:complete."

W

Watchdog -- A monitoring component that detects stalled agents (no output after a timeout) and marks them as failed.

Workspace -- An execution environment with a specific configuration: local path, SSH connection, or Docker container. Agents in a chain run within a workspace.

---

Missing a term? Let us know. Building your first chain? Start here.