Glossary

The terms used throughout the playbook. Precise words lead to precise designs.

Agent: An AI worker configured in raia cX, with instructions, a model, and optionally knowledge, skills, channels and tools.
Agent access: A connector's live job: exposing selected tools in another system to an agent, usually over MCP through a bridge in Command.
Approval: A Loop step where the run pauses for one or more people to approve or reject. A rejection can send the work back for revision.
Bridge: A connection managed in raia Command that handles sign-in to another system and gives an agent a scoped set of live tools.
Connector: A link from raia Command to a system you already run. It can import content, give agents live access, or both.
Conversation: An ongoing, multi-message exchange with an agent, visible in Copilot.
Escalation: A decision made during a conversation to alert people and, optionally, hand the conversation to a person.
FTEE: Full-time employee equivalent: the hours of capacity agents return to a team each month, measured from their activity in raia.
Import: A connector's content job: copying relatively stable material into a Command knowledge folder for training.
Knowledge: Curated reference material in raia Command, prepared and trained into an agent's vector store.
Loop: A versioned workflow in raia Command that starts from a trigger and runs steps, branches and approvals in order.
Loop AI: A Loop step that reshapes, summarizes, classifies or scores earlier results without an agent and without taking actions in other systems.
MCP: Model Context Protocol, the standard raia uses to connect agents to tools in other systems, and to let assistants such as Claude reach raia agents.
Orchestrator: A skill that lets one agent ask specialist agents for help and combine their answers, without handing over the conversation.
Prompt: A single task sent to an agent, as opposed to an ongoing conversation. Loops and orchestrator calls use prompts.
raia Control: The product for outbound campaigns over contact lists. It's separate from a single message sent by an agent or a Loop.
raia MCP: The setting that exposes a raia agent as an MCP server, so other assistants can call it. It's the opposite direction from giving an agent MCP integrations.
Run: One recorded execution of a Loop, with a permanent history of what happened.
Specialist gate: A design where a light front-line agent passes consequential actions to a specialist that checks the details before acting.
Tables: Structured rows and columns in raia Command, with read-only agent access, for exact lookups that knowledge search can't provide.
Vector store: The search index behind an agent's knowledge. It finds relevant passages; it isn't a database.
Webhook: A message sent from one system to another when an event happens, such as a conversation ending. Webhooks often start Loops.