Key Concepts

Use this glossary to distinguish the parts of SquadOS before configuring a journey. Names in bold match the English interface.

Organization, members, and access

An organization is the isolated environment that contains agents, bases, tools, inboxes, contacts, conversations, automations, and settings. A person can belong to more than one organization and switch between them; data and permissions always belong to the active organization.

Each member receives one or more Roles. Their capabilities are combined and determine which pages and actions the person can access. The current packages are Administrator, AI Agents, Support, and Collaborator; Custom lets you select capabilities for one specific person. The Owner is unique, has full access, and is not a selectable role.

Teams are the legacy model for some organizations. In that model, they also restrict agent categories and model options. Do not treat Team and Role as synonyms. See Users, Roles, and Teams.

AI Agents

An AI Agent combines:

• a prompt with its goal, context, and rules;
• a language model;
• response, security, and attachment-processing settings;
• optionally, bases, context blocks, and tools.

The agent uses only the configuration that has been saved. A draft open in the editor does not yet change chats, Inboxes, or automations. Deactivating the agent preserves its configuration but prevents new responses in channels and automations. See Create AI Agents.

Knowledge

A Knowledge Base is a collection of items the agent can search. Files, texts, crawled pages, and other items must finish processing before they become searchable. Connecting a base to an agent makes the search tool available, but does not force a query on every message: the runtime and model decide when to call it based on the request and the agent’s instructions.

A Context Block is different: when linked and active, its full text is sent with every agent response. Use blocks for short rules and facts that must always be present; use bases for larger collections that should be searched on demand.

AutoLearn records questions the agent could not answer from its bases. An opportunity becomes content only after human review and approval; it does not train the model or publish knowledge on its own. See Knowledge Bases.

Tools

A tool lets the agent query data or perform an action. The organization’s catalog combines four sources:

• native SquadOS actions;
• Composio integrations;
• custom HTTP endpoints;
• tools discovered from MCP servers.

Making a tool available in the catalog does not connect it to every agent automatically. It must be added to an agent and, where applicable, have its actions and credentials configured. During a conversation, the model decides whether to call it based on its name, description, parameters, and the prompt. A connected tool therefore does not mean that it runs on every message. See Organization tools.

Inboxes and channels

An Inbox is the configured entry point for an external channel. It stores the required connection and determines whether new messages go to an AI Agent or human support. Routing can also vary by schedule, date exceptions, and the destination outside business hours.

Current types include WhatsApp QR Code, Z-API, Uazapi, Evolution, and Official WhatsApp; Telegram; Instagram Direct and Comments; Public page; Widget; and API. One agent can receive messages from several inboxes, and an inbox can work without an agent when its destination is human.

Inboxes are managed under Inboxes, outside the agent editor. The documentation’s historical URL section still uses “triggers,” but the current interface name for channel entry points is Inboxes. See Inboxes.

Contacts and conversations

A Contact represents the person being served and brings together their identity, attributes, tags, and related history within the organization.

A Conversation is the message thread received through an inbox or channel. It has its own operational state, owner, contact, agent, and AI control. On the Conversations screen, the team can claim, transfer, pause, resolve, or reopen service according to their capabilities. Turning AI off in one conversation does not deactivate the entire agent.

In simple terms: the inbox receives, routing chooses the destination, the conversation records the service, and the contact identifies the person. See Manage conversations.

Hub

The Hub is the internal, authenticated workspace for organization members. There, a person can chat directly with AI Agents and, when they have access, use Rooms, direct messages, and meetings.

A Hub chat is not an external service queue and does not appear as an Inbox. Do not use the panel’s Conversations screen to look for an internal Hub chat. What each member sees depends on their capabilities and access to agent categories. See Hub overview.

Automations

An Automation is a flow that begins with an event or time trigger and moves through connected conditions and actions. It can send messages, update data, wait, branch, or call AI resources, depending on the configured nodes.

While you edit, the graph is still a local draft. Save validates the graph and creates a new version that is already published; there is no second publish action. The Active switch is separate: for an active automation, the saved version applies to new runs; for an inactive automation, it remains stored until activation. Runs already in progress continue on the previous version. Results appear in the automation history. See Automations.

AgentMaker

AgentMaker is the editor’s AI assistant. In Agent mode, it analyzes the saved configuration and can propose changes to the prompt and tools. In Automation mode, it helps build or review the flow graph.

Proposals are shown for review. Applying one changes the form or graph being edited, but does not replace the normal save step. See AgentMaker.

How the pieces connect

• External service: channel → Inbox → routing → AI Agent or person → Conversation linked to the Contact.
• Internal use: authenticated member → Hub → AI Agent, Room, direct message, or meeting.
• Automatic execution: event or schedule → published Automation → conditions and actions.
• Agent response: saved configuration → model → applicable context → tools called when needed → response.

These flows can intersect, but they keep their own controls. For example, an automation can start an action related to a contact, and a tool can transfer a conversation; that does not turn the automation into an Inbox or the Hub into a service queue.