Core concepts
- Squad: a set of two or more agents with a shared service purpose.
- Each agent in the squad maintains its own LLM model configuration, personality, and tools.
- The limit is 20 active agents per squad.
- A channel can only be linked to one squad at a time.
- Agent role in the squad: defines the function the agent performs within the group.
manager— leads routing in hierarchical mode. Central triage point.principal— coordinates without rigid hierarchy in horizontal mode.member— specialist agent that receives delegated or routed tasks.specialist— member variant with declared competencies for criteria-based routing.
- Handoff: transfer of a conversation from one agent to another. Logged with source agent, destination, and reason.
- Squad ID: unique identifier of the squad, used when linking channels programmatically via API.
Operation modes
The mode determines how routing happens within the squad. The choice is made at creation and affects the required role structure.- Hierarchical (
hierarchical):- A
manageragent receives all messages and decides which member to forward to. - The end user interacts with a single entry point — triage happens behind the scenes.
- Requires exactly one
managerto be defined before the squad is activated. - Ideal for: support with specialized queues, service with gradual escalation.
- A
- Horizontal (
horizontal):- A
principalagent coordinates, but all members can be triggered directly. - Routing uses criteria such as language, channel, time, or contact tags.
- Requires exactly one
principalto be defined before the squad is activated. - Ideal for: teams with similar competencies, load distribution by criteria.
- A
Why use squads?
- Resolve complex conversations that require more than one specialty without scaling the prompt of a single agent.
- Create single entry points per channel while multiple specialists work in parallel.
- Log and audit every handoff, identifying bottlenecks in the service flow with real data.